September 23, 2021 • 50 mins
We are excited to have Mike Polacek as a guest of honor on this episode. Mike is one of the few business leaders who has run large tech businesses and has been a professional investor in public companies.
We dig deeper into the how building semiconductor business is different than building a SaaS business, complexity, Moore's law, and many other interesting subjects related to tech businesses.
I enjoyed my conversation with Mike and hope that you will enjoy it as well.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Chander Chawla (00:00):
Hello everyone.
(00:00):
Welcome back. This is ValleyNordic a podcast about culture,
(00:02):
technology and business from twoperspectives. That's my
perspective from Silicon Valleyand arne's perspective from the
Nordic countries. But today wehave a special guest, you'll get
two Silicon Valley perspectives.
(00:24):
I am with Mike politek InPescadero. That's the place
where Gordon Moore was born andraised. You. I'm sure I've heard
of Moore's law. So that GordonMoore was raised here, and Mike
politek is one of the smartestguys I know, and I had the
(00:45):
privilege of working with him,and I would say he I usually
like running my own show, butI'm happy to be his number two.
Anywhere you it's hard to findpeople who have the intellectual
depth and also who are ability,you know, their their ability to
communicate, you know, verycomplex idea in simple terms,
(01:07):
and align lot of people. So youwill see that today. So Mike,
welcome to Valley Nordic, andthanks for joining us. Thanks,
Mike Polacek (01:14):
Chander, that's introduction is too kind. And
also, I'm glad to do this faceto face, I don't know, for
breaking any rules, any laws.
Hopefully not we can edit thisout.
Chander Chawla (01:26):
We are both vaccinated, and we are in Mike's
beautiful place in Pescadero,and we're six feet apart. We are
six feet apart, and Mike has avery and I think everybody who
is a regular to valley Nordicknows my affinity towards
Berkshire, Hathaway to Buffettand Munger. So I see Mike is one
(01:49):
of the few executives who hasbeen on both sides, meaning he
ran big divisions. He waspresident at Knowles. Before
that, he was SVP at NationalSemiconductor, but so he has run
large businesses, created newbusinesses, and he spent four or
(02:09):
five years on the investingside. So he he's an operator and
also an investor. So which youknow, if, from the startup
perspective, the biggestchallenge you have is the
capital allocation. So you dothat internally, within a
company as a CEO or a startupfounder, but from externally,
(02:33):
you do you spread it apart toother companies, so you have
that perspective. So I find thatvery interesting, and today
we'll discuss, we'll do a deepdive into semiconductors. You
know what they are, why they'reimportant, and basically past,
present and future. And Mike,you want to say something before
(02:55):
I dive I'm so eager to dive intoit. But if there's anything you
want to add to the intro? No,
Mike Polacek (03:01):
I would just say we could definitely dive in that
your comment about investing andportfolio management, of your
investments and capitalallocation, resource allocation
within a company is more or lessthe same sort of idea as if
you're doing it outside, eitherinvesting in a division or
something you want to buy, ifyou're a strategic investor, or
(03:22):
if you're just a public marketinvestor, investing in stocks,
it's sort of the same thing thatthe diligence that you do in
terms of evaluating the productsand the technology and the
management team, and thenultimately valuation and all the
rest of it is, you know, it'sthe Same, same general concept.
And so I found that doing thatkind of work inside national
(03:45):
was, you know, quite applicableto doing it outside as
professional investor.
Chander Chawla (03:49):
Yeah, great. So I want to start with the
fundamentals of siliconbusiness. Because, you know, a
lot of the training people getin the startup world. When these
days, when people think ofSilicon Valley, they think of
basically B to B, SaaS startupsor consumer companies like
Facebook, Google or slack wouldbe B to B, example. But it
(04:15):
started, it's called SiliconValley because of
semiconductors. So how are thein your view, the fundamentals
of silicon business differentthan, let's say, A, B to B, SAS
startup.
Mike Polacek (04:29):
Well, I think the most obvious first thing I can
think of is just the investmentrequired and the time frame
required to get a product out.
And you know, it's verydifficult to sort of spend a
little bit of money testsomething, see if it works,
adjust, modify, get somethingelse out incrementally, a little
(04:49):
bit better. You know, that'sjust not how semiconductors
work. Generally speaking, youmake a fairly large investment
depends on the complexity of thechip and sort of. What you're
working on, but, but you canmake a multi year investment, a
multi per multi 100 person yearinvestment, 10s of millions of
dollars investment, before youcan try the product out and and
(05:12):
that that requires, I would say,a lot more discipline. There's a
lot more risk, but then you needa lot more discipline up front
and and then you need to, youneed to hit big, big markets,
because those investments arelarge and the risk is large. So
so then you better be, you know,fairly sure that you're going to
(05:34):
be able to address a market thathas the right number of zeros
for that to have been a goodinvestment,
Chander Chawla (05:42):
and how do you know that that market is there?
Mike Polacek (05:47):
Well, you don't know. I mean, I would say you
could invest in a market thatyou're already in. So if you're
already in a market and you'resort of doing incremental
improvements, then the risk islower. You you may have gotten
this feedback from yourcustomer, or your customers
customer, that this next thingthat you should do, that there's
(06:07):
a demand for that, even evensaying that, you know, even if a
customer comes to you, whichhappens all the time, a big
customer will come to you andsay, oh, I want this next
version of this last thing youhave. But guess what the that
customer is saying the samething to your competitors, yeah.
So you're racing to try to getthat product out, to sort of
(06:27):
meet the performancerequirements, get the cost
structure in place so you canhit price targets, get the time
frame right, and you do all thatinvestment for that customer
that says, I want this or a setof customers. And then you get
there and you deliver samples,and you find out that, well,
your competitor did it a littlebit better, or a little bit
cheaper, or whatever, and nowyou're sort of in the back seat
(06:51):
at best. So there's still riskthere. The other opportunity is
to go after a market that's new,which is riskier in some ways,
in terms of, you know, is thatmarket really going to be there
when you arrive two years, threeyears later, at the size that
you need it to be? And then, butthen the risk goes down,
necessarily, or maybe notnecessarily, but the risk
(07:12):
certainly goes down with respectto competition. So it's not as
obvious a market. Maybe, maybeyou've seen trends that other
that you've maybe yourcompetitors haven't seen. And so
then, when you arrive at thatmarket, and that market's there,
there may not be as manycompetitors.
Chander Chawla (07:29):
It's, I remember you told me this, and I still
remember this, because it madean impression. You know, LED
lights, the first commercialavailability was in like, 65
Mike Polacek (07:41):
Well, the, yeah, the invention, you know, it was
a University of Illinoisprofessor that my alma mater,
and it was, it was decadesbefore I was there, and I'm not
young, that he invented LEDs.
And he actually made aprediction about that someday
that LEDs would be used forlighting instead of incandescent
lights. And, yeah, it took manyyears, you know. And that's, you
(08:04):
know, I always say that thebigger challenge in identifying
new markets, at least for me, myexperience, has been, for my
career, has been the timing. Youknow, more than Are you right or
not right? You know, I wouldsay, yeah, you can look at LEDs
in the 1950s or 1960s and say,Yeah, this is a great technology
(08:25):
that emits light, and it'sinherently lower power, and it's
but okay, there's a bunch ofthings that have to fall into
place for them to becommercially available. So when,
if you're sitting there in 1950or 1960 you know, write down
what's the year you think LEDswill become popular, you'll be
able to buy them at whateverhardware store. And, you know,
being able to do that accuratelyis very difficult, yeah, I
(08:48):
remember, you know, in selfdriving cars. So this was, you
know, I think it was maybe 567,years ago, when they were first
starting to kind of get into thedialog of, of, okay, we might
have self driving cars and andthere's different versions and
different levels, and maybe, youknow, trucks and busses and all
(09:11):
these so. So this was, like abig thing in 2000 I don't know,
1415, something like that. And Iremember reading or listening to
and I hope I get this right ofthe seats. I think there's a CEO
Ford, or some high levelexecutive at Ford coming out and
saying, you know, by in the nextfive years, this was in 2015 or
(09:33):
2016 so basically, by today,2021 Ford will be making cars
without acceleration,accelerators, pedals, steering
wheels. That was going to bemainstream, mainstream enough
that Ford's good, right? Thisis, this is, you know, and I
remember reading that thinking,here's a guy that just doesn't
(09:56):
understand the technology,right? But I'm sure he's very
smart, very accomplished, but hegot that wrong. Now, it doesn't
mean that that's never going tohappen, that that's an absurd
idea. It's like a crazy he's notbreaking laws of physics by
saying that he's he's justreally wrong with respect to the
timing, and that's, I would say,really wrong related to the
(10:18):
complexity of the problem? Yeah,and I think you just
underestimated the complexity ofthe problem. And I think you see
that kind of thing in a lot ofplaces, in a lot of markets,
where it's sort of obvious, thisis going to happen, but where
you really need to be right isin the level of detail about how
is it going to happen, when isit going to happen? Where is
(10:39):
going to happen if you're reallygoing to exploit a big trend
that's maybe fairly obvious, youknow, you need to dig several
layers down. And I would saythat's a much bigger challenge.
Chander Chawla (10:50):
Is there a method or system people can
follow to do that?
Mike Polacek (10:56):
I don't know. I I think there's different phases
of trying to identify newmarkets. So, so one might be
like a brainstorming phase. Sowhat's possible? You know,
linear thinking is problematic,I would say, to being creative.
Chander Chawla (11:15):
Say more that's interesting, like, say more on
why linear thinking isproblematic.
Mike Polacek (11:20):
So I mean linear thinking will get you linear
results, so it'll get you thenext thing like and that may be
great. There's nothing wrongwith doing the best next thing,
and that's an aspect ofinnovation, for sure. But if you
want to do something a littlebit bigger impact than that,
then generally those things arenonlinear, meaning they don't
(11:41):
follow that. Okay, six monthsago, this improved this much, so
six months from now, this isgoing to improve this much,
like, that's linear thinking.
And most and humans, I wouldsay, in general, are pretty
linear thinkers. Like, it's justsort of, I don't know it's in
our DNA or hardware wired, likewe don't react well to, you
know, exponential curves, like,we sort of can't get our heads
(12:03):
around that. And you see thateverywhere in society, people
just don't, sort of can't makethe connection between, what's
the difference between a billionand a trillion, or what's the,
you know, the one of these gamesI remember is, is, would you
rather have a million dollars.
I'll give you two options. I'llgive you a million dollars
(12:24):
today, or I'll give you a pennytoday and I'll double it every
day for 30 days, right? Turnsout, if you take a penny today
and you double it every day, onepenny one day, two pennies the
next day, on and on that thatsecond option would be a much
bigger number than the firsttime, right? It's just hard to
think like, yeah. You know, it'sjust people don't think that
(12:47):
way, yeah. So, so in order to bea non linear thinker, you have
to get out of that trap. And howdo you get out of that trap? I
think there's certain, you know,there's certain brainstorming
exercises you could do that are,you know, to try to try to get
yourself, you know, one is, isthis? Who's listening to this
(13:09):
guy? I think his name is deBono. Edward de Bono, if I have
that right, he's, he's got thesesix hats kind of thing. And Sri
has used this, I think Sri hasused this methodology. They have
versions of this where they yousort of play different roles.
You get a group of peoplebrainstorming, and people are
sort of play different roles toattack and to, you know, to and
you sort of get out of, try toplay, you literally play roles
(13:32):
to try to get out of theconventional thinking. You know,
there's, there's, there's thingsof looking for convergence,
right? Looking for two marketsthat exist and that are
happening and that you're sureare going to happen. And then
you just sort of write themdown, put them next to each
other, and say, it can be stupidstuff, right? It could, but, but
(13:52):
there could be some, some goodthings, right? So you take, you
take 5g which is like a bigobvious thing, and you take IoT,
which is a big obvious thing,and then you stick them together
and and then you sort ofbrainstorm what could happen if
you bring 5g kind ofconnectivity to a Iot, type of
(14:13):
sensor network world, right?
Okay, there could be all kindsof cool things can happen. You
could imagine and on and on andon. You could look for gaps, you
know, in in technology, whereyou sort of hit a wall and then
something, you know, what couldsort of extend that wall. So,
yeah, you could look for, youknow, all kinds of different
(14:37):
ways to kind of get yourself inthis creative and non linear
sort of thinking world. Yeah,
Chander Chawla (14:43):
it's hard, because you're right. Most
people think that way. And youthink about what's taught in the
software world, in the MVP, andhow to take products to market,
that's what's taught. You know,you create an MVP that's.
Minimum viable product, and youget it out in the market as fast
(15:04):
you can. And then you iterate,yeah, you know, get feedback,
which makes sense, it becauseit's software, but like you
shared in the very beginning,the fundamentals of the
semiconductor business aredifferent. You can't really
iterate your way to success. Imean, unless you have billions
of dollars, you can iterate, butas a startup, it's hard to
(15:26):
iterate your way to success withlinear thinking or incremental
improvement and capture anythingbig. Yeah,
Mike Polacek (15:35):
and, you know, I mean lots. There's, again,
nothing wrong with incrementalthinking and software or
anything else. I mean, a lot ofthe I think people also
overestimate, sort of like anindividual God, like creator
that invented, magically, theiPhone or the or the PC or the
(15:56):
operating system like the like,if you know the details of
these, Every one of thesestories there, it's, there's
nothing sexy about like, it's,it's 1000 small things come
together to create a greatproduct. And, and it's maybe
that attention to detail ofbringing them together, and the
attention to detail of executingon a schedule. And, you know,
(16:17):
those kinds of things which arenot sexy. You know, read them in
the big books by,
Chander Chawla (16:20):
you know, Walter Isaac by Walter Isaacson,
Mike Polacek (16:23):
exactly, so, but, but that's usually, you know,
the story behind the iPhone, orthe story behind, you know,
Microsoft OS, or something like,it's not, you know, there's,
there's not, usually, now,sometimes there is magic, but
Oftentimes the magic isaccidental, yeah, which? Which
is also this nonlinear sort ofthing. A lot of like, the
(16:45):
coolest inventions areliterally, like, lab accidents
and things like that. Where,yeah, that's, how do you
capture? How do you make sureyou're ready for the accident
when it happens? You know,that's also a mindset. If I see
something that's a surprise thathappens, what's my reaction? Is
my reaction, oh, let's throwthat in the garbage because it's
a surprise. Or is my reaction,oh, my God. Like, this is really
(17:08):
cool. Like this, this is, thiswasn't supposed to be this way.
Let's, let's, let's, let's,let's do some more work on that.
Chander Chawla (17:13):
Yeah, yeah.
It's, can we do two minutessidebar on the accidents? You're
right. Lot of them are accidentswith things moving to remote
work and hybrid work. What doyou think happens to those
accidents?
Mike Polacek (17:32):
Well, some, some can be done by individuals, I
suppose, like a lab, a lab.
What's a lab in the SAS world?
It's maybe somebody's laptop intheir kitchen, right? I mean, I
I'd have to think about it somemore, but, but, I mean, I would
say, certainly one of the thingsyou miss is the serendipity of
(17:53):
human brainstorming, person toperson interaction, where you
have an idea and you bump intosomebody. I bump into you. I
said, Hey, I was just thinkingabout this. I had this weird
dream, and I wrote down thesenotes at two in the morning, and
I and I woke up this morning andit's like, oh my god, this is a
great idea. What do you thinklike? But is it crazy? And then
I go, you, and I say, Oh, isthat crazy? And you say, well,
(18:14):
it's a little crazy, but let'sask Jim. He's right over here.
He's an expert on this,whatever, blah, blah, blah. So
that kind of thing happens,right? That kind of thing
happens a lot, where you cansort of iterate, brainstorm,
bounce things off each other.
You know, the barrier that's notimpossible to do in a sort of
virtual environment, but thebarrier to do that is higher,
(18:34):
like the am I going to call youor I'm going to schedule a zoom
call with you, because I hadthis weird dream. I would say,
No, it's a little weird. Iprobably wouldn't do that,
right, but I might just mentionit to you by the coffee machine.
Yeah.
Chander Chawla (18:50):
I mean, generally, I found if whenever
we had these conversations orwith other people, they're more
in casual environments, we arehaving coffee, or we are having
a beer, or like, becausenormally during the work day,
you're so focused on what youneed to deliver the quarterly
(19:11):
results, or whatever it is, thisis more when you're in a
different mind frame and that Ifind it hard. You know how we'll
see how it get it getsintegrated in the remote work
world,
Mike Polacek (19:22):
yeah? I mean, it's certainly more fun to do this
kind of thing face to face, youknow, versus, you know, sitting
around by yourself, working ontasks all day, yeah? Which is,
you know,
Chander Chawla (19:35):
okay, let's come back to semiconductors. Can you
like, I always found thisfascinating, the value
distribution, you know, atnational we were making these
chips, which were, you know,like, basically, you couldn't,
let's say, ship a phone withouta PMU. But PMU is power
(19:58):
management unit that. Helps youknow how the power to distribute
it to different parts in thephone. So if that was, maybe we
can share it like sub $2 parts,and the actual bomb or bill of
materials of the phone was atthat time, let's say, 200 bucks.
(20:20):
So that's like, 1% and like Ifelt everybody we are doing, you
know, let's go back to the firstthing you said, how much risk it
takes, how much investment ittakes to get things right. So
the semi people are taking allthis risk, failing multiple
times, to get these thingsright, and making investments,
(20:43):
and then it goes to the phonemanufacturer, which makes huge
margins, lot of money. Then theoperator, who sells the phone or
has a service plan, they need alot of money. So the value
that's created in thisecosystem, how is it that the
guys were doing, or gals, thetaking the most risk, are
(21:04):
getting least reward.
Mike Polacek (21:07):
Well, I would say the margins on those devices
were high, probably higher thanthe phone guys were making. So
that's one indicator that thatthere is value being placed on
it. So maybe not value in termsof margin dollars per phone, but
in terms of margin percent, itwas high. I think the risk part
(21:28):
of it is, you know, it's aconscious decision. If you're
going to go after a socket in aphone, it's you're making a
conscious decision that I'mgoing to go after a market where
the power dynamic because of thevolumes of the phones and the
concentration of phone makersversus the concentration of
suppliers. That I'm making aconscious decision, that I'm I'm
(21:52):
taking on a risk there. I'mwilling to, I'm going to be
playing a subservient role wherethis guy is good, this this
phone guy is going to push me tobe able to push me around if he
wants to. And, but why do that?
Well, you do that because,because that Phone Guy might be
shipping 100 million units ayear. You know, tell what other
market could you? Could you, youknow, for a couple bucks, could
you get into a market thatyou're that that one guy could
(22:13):
give you a design win that's,that's 100 million units. You
know, it just doesn't exist. Youknow, you get excited about a
design win in a in a car or a TVor whatever, you'd be lucky to
ship a million units, right? So,so now, my personal opinion is
that that's a that's date like,it's very dangerous. It's a very
potentially is a big, bigmistake. And there's a lot of
(22:36):
companies around, startupsaround, that no longer exist
because they were taken out bythese companies. That these
companies are prettybloodthirsty, you know, they'll,
they'll sell your theirgrandmother for a nickel. And so
if you, if a competitor comesalong, and oftentimes, these
phone guys will, will guide yourcompetitor, if they feel like
(22:58):
you're you have a little toomuch power though, maybe somehow
the planets aligned and you yougot something that no one else
could do. And maybe you want tocharge a little more. You want
to do whatever the Phone Guysays, not in my market. So
they'll help their competitorand then literally take them out
of business. And that's, youknow, if that's your only
(23:20):
customer, you get very excited.
You get 100 million unit deal.
You go from zero to $200 millionin revenue. All of a sudden,
champagne corks, maybe you gocrazy, and you want to go
public, you do all these kind ofthings. And then, and then, all
of a sudden, that that's acustomer says, oh, you know, I
really don't like that productanymore. You should cut the
(23:41):
price in half, or I'm gonna usesomething else. And you're like,
what? And then you implode, andyou have to fire everybody, and
you cry. You don't have yourselfto sleep. So So I would say,
Yeah, you should go into thatwith your eyes open. And there's
certainly other markets whichare smaller volume where that
power dynamic isn't the same. Infact, the same. In fact, the
power dynamic may be theopposite. The power dynamic may
(24:03):
be that you know you're youbecome such an important partner
to that customer that they treatyou with respect and longevity
and loyalty and all these. Andyou could do it mutually. And so
now you won't ship 100 millionunits in a year, but it's sort
of, you know, what kind oftrajectory Are you looking for,
and how stable a business areyou looking for? So these are
(24:25):
conscious decisions. I would sayneither is right or wrong. If
you're in a huge company likenational or some other company
where we had 1000s, literally1000s of products, you know,
maybe a dozen, couple dozendifferent product lines. Yeah,
you can have a portfolioapproach where some of them,
some of my investments, aretowards this very high risk, but
(24:47):
potentially home run typevolume, and then other
investments are on the otherextreme, and lots of between.
And I would say that would be asmart way to deploy your assets.
Chander Chawla (24:59):
Yeah. It's interesting. You know how that
happens? You think the cycle,the phone replacement cycle, has
something to do with it?
Because, you know, peoplegenerally replace their
smartphones every two years, orand companies release their
phone every year, so they have achance to change things every
(25:19):
year. I remember when I workedwith Siemens, you know, you were
selling network equipment, sorouters and big switches and
DWDM stuff, and they it's intheir network. If something
fails, they lose business. Sothey always wanted three vendors
(25:39):
for everything they did. So ifsomething fails, but the
relationship was there, theycouldn't just throw you out,
like in the phone business, theycan just throw you out next year
here that you're integral to howthey make money here, that cycle
changes every year. Do you thinkthat has something to do with
it? Because the fast cycle?
(26:01):
Well,
Mike Polacek (26:02):
I mean, the consumer market definitely has
less of a requirement forloyalty than other markets that
are longer cycle markets. So Ithink that's right, you know,
will it? I mean, you know,replacement cycles are
lengthening in phones, right? Itused to be 20 months, and then
it was 22 then it was 2426 28 Ithink it's probably getting
(26:22):
close to two and a half yearsnow, instead of two years. And
so, you know, will that breedmore loyalty? I doubt it. You
know, it's, it's not somethingthat you know. I think it's, I
think it's the volume you know,more than the, than the cycles
(26:45):
where, where the, just thevolume is so high. And every
penny, half Penny, you know, Isaw, I saw this. I mean, it's
some at pretty much all thesecompanies that I've worked with
that were good customers, thatsort of became not so good
customers, you know, we saw itwith IBM in the literally 80s
and 90s, saw it with Nokia,where we used to have meetings
(27:07):
with the design engineers, likewe would be semiconductor guys,
we'd have all this greattechnology we were thinking
about working on, and then we'dhave the design guys for the PC,
and they were talking about whatthey needed, if they could dream
of something, what would it be?
And we would write it down, we'dcome back a month later, and we
would have those kind ofconversations, and and then the
(27:27):
planets would align, and then atsome point, you would come and
you would have developedsomething, and you'd give them
samples, and they would use it.
And everybody loved each other,right? And, and they got a
great, innovative solution outof it. They got something then
that they could do somethingthat their competitors couldn't
do. And, you know, we did thesame thing. Yeah.
Chander Chawla (27:54):
So the I think Mike has spent a lot of time in
the industry I spent limitedtime so you're seeing how now
he's talking about how thingshave evolved, how it used to be,
and then what it became afterthe customer's success or market
position changed, yeah,
Mike Polacek (28:14):
the volume got bigger. So then what I saw is,
then slowly, what happened isthat the IBM every meeting, then
at some point, the procurementguys would be there. And so then
it would be like in the in sortof product engineering guys. So
then you would have like, halfthe people in the room were
design and design guys, and thenthe other half were sort of
more, sort of plumbing guys. Andthen a year later, and then
(28:37):
there'd be like one design guy,and he wasn't allowed to ask
questions, you know, and it wasall the procurement and product
guys and then, and then, in theend, like you couldn't, you
know, you're later, like youcouldn't meet the zeitgeist like
the other guys, they just wantedto talk about your existing
products and what you're doingto cost reduce them, and what
you're doing. And maybe theproduct guy would say, you know,
I need something with slightlylower power dissipation or
(28:58):
slightly higher performance,whatever. I'm being dramatic.
But it was sort of that story.
And that same story happenedwith Nokia. I remember Nokia, we
used to have these greatconversations with Nokia, these
design engineers like, fly allover the place, and we would go
there, and they would come hereand and there'd be all this
designer to designer type, greatstuff happening, technical, deep
technical conversation aboutwhat's possible, like, and
that's the magical interfacebetween, like, what's possible
(29:21):
and what's needed, yeah, right.
And so you both togetherdiscover that, and then we would
do these investments and do thiswork. And it wouldn't be
foolproof, but it would, yeah,you would create really cool
stuff out of that. And, but, butthen Nokia became this giant
phone company, and thosemeetings changed. You know that,
(29:43):
you know, and I think, fromtheir side, why do they change?
And their side, they're like,well, we don't want any like
silicon supplier to get too muchpower. So we don't want you,
like, in the back rooms with ourtechnical people to sort of
position me the procurement. Guyinto a sole source position
where I really want somethingyou have, and you're the only
(30:04):
guy that can supply it. Sothey're trying to prevent that
from happening where, where youhave this really important,
unique product that they need,right? And it's sort of built
into the system. And then I sawit, you know, again, you know,
more recently with companieslike Apple, that is sort of the
same thing. I remember apple. Weused to have these great
(30:25):
conversations with Appledesigners, and now I don't
million years. You don't get toget the same building to ask
people
Chander Chawla (30:34):
that, I mean logically. Well, it depends what
logic you look at. So what youjust described the procurement
team or bigger companies aredeliberately saying, we don't
want to innovate, becauseinnovation in the ecosystem
comes from one initially, fromone company. So one company is
(30:55):
going to have a breakthrough,and you can have access to that.
But what, when the companies getbigger, it becomes, you know,
risk avoidance, rather thancoming to a new thing, you can
get first and have more revenue.
Is that? Do I get that? Yeah,
Mike Polacek (31:13):
I think it's, what do you emphasize? What's the
most important thing to you? Andat some point, the volume gets
so big that what becomes mostimportant to you, and this may
be totally correct, like forshareholder value or the look at
Apple. Apple's obviously beenunbelievably successful under
Tim Cook, but the emphasis hasbeen more on production,
(31:34):
reliable manufacturing, gettingmargins stable, being able to
ship in unbelievable volumes atreliably high quality. All those
kinds of things have become moreimportant than innovation. I
Chander Chawla (31:50):
would say, I would say, everything you said
is correct, but innovation,they're doing internally.
They're doing verticalintegration. So they're making
their own chips. They're, youknow, influencing Corning. So
they there's lot of innovationstill happening, but it all in
house. So they've reduced thedependency. They still want to
(32:13):
innovate. And they may havenoticed what you just described,
procurement people, or whateverreason you can't really get
innovation in, in, in the doorfast enough. So let's do it
internally, in,
Mike Polacek (32:25):
yeah, yeah. And I'm sure there's a lot of great
innovation happening there, Ithink, but, but the The old
saying is true still, that mostof the great engineers in the
world don't work for fill in theblank company. Yeah. If you're
sitting at Apple, you have to behumble and say, you know, most
of the great engineers in theworld don't work at Apple. You
(32:46):
know, the vast majority don'twork at Apple. No, that doesn't
mean Apple doesn't have greatengineers. Of course they do.
But if you can put two companiesside by side, one company that's
doing everything internally, andone company that's able to sort
of create a nurturing ecosystem,a bunch of people that are doing
that, they're doing work forinternally, but also doing work
for them externally. I couldtell you which company will be
(33:09):
successful long term, and, and,or at least, ultimately
successful, or most successful,
Chander Chawla (33:16):
it's like, let's take the Samsung example. You
think they're doing internal,external combo, or like they
compete with Apple, but Apple, Iwould say, still is, you know, I
Mike Polacek (33:29):
mean, there's been a huge trend. I mean, for a long
time during my career, there wasa trend all these system
companies were all putting theirsemiconductor stuff out, like
we're spinning them out, right?
Every single big company,Phillips, Siemens, IBM, right,
they were all spinning out. Why?
Because the capital investmentswere so large and and they were
(33:51):
getting larger every generation.
The factories were costing more,the development costs more. And
it was just like, Okay, we don'thave the volume to so we're
going to spit this thing out.
And that was going on fordecades, and definitely in the
last whatever five, 510, years,it's definitely been moving back
in, where you get more verticalintegration. Samsung's been
(34:13):
doing it. Huawei has been doingit. Apple does it. Even guys
like Google and Amazon you don'tthink of as hardware companies,
you know, have their own hugechip development, yeah,
organizations now. So, you know,why is that? You know? I would
say there's a few things, youknow, one which we could talk
(34:35):
about, if you want, now orlater, but sort of, what's the
impact of the of the slowdown orelimination of Moore's law?
Chander Chawla (34:45):
Yeah. Let's talk about it. Yeah. I think very
interesting, yeah. I mean,Gordon Moore's, you know,
territory. Let's talk about it.
Yeah, it's,
Mike Polacek (34:55):
you know, first, when I say Moore's law, what I
mean it so I think people talkabout it when you. Read things,
it's sort of, I think it gets alittle bit muddy, but what, what
you see is, can you still bruteforce shrink transistors? If
that's what you think of Moore'slaw, then, yeah, I think you can
still shrink transistors at somepoint. There's an end there. But
(35:16):
they're they're gone below 10,and they're going to seven, and
then people talk about three,and where's the wall, and you
get x ray lithography and gammaray, I don't know, all the sort
of you can make light smallerand smaller and and, you know,
so that part, I would say, Okay,I'm not even gonna argue that
maybe you can still shrinktransistors. Fantastic. But the
magic of Moore's Law was theeconomic side, was that you
(35:41):
could shrink transistors andmake transistors or gates
cheaper by a lot everygeneration.
Chander Chawla (35:49):
Yeah, that's around 18 months to two year
yeah, and,
Mike Polacek (35:54):
and, and that's the magic of a lot of markets.
So if you're a system company,and you know, every 18 months, I
can get something that's goingto vastly outperform the last
thing at the same price, thenyou can make a system around
that that's going to do the samething, and then you're going to
create demand for that newproduct, for that reason,
(36:16):
because now your new systemproduct is going to do some
wonderful things it didn't DoBefore at the same price, right?
Or maybe, if the other side is,you can do it at, you know, the
same performance, you can do ita lot cheaper, yeah, which is
another sort of dimension, yeah,and, and that's the magic of
Moore's law. And I would say, Idon't maybe this is hyperbolic,
because I'm a semiconductor guy,but I think that is the biggest
(36:40):
economic impact for the UnitedStates in my lifetime, that that
so many industries are sustainedby that concept of having the
next thing be so much better atthe same price every 18 Months.
Think about a world where thatdoesn't happen, where, where,
(37:02):
where your your choice, 18months later is to buy something
that's more or less the same atthe same price. What are you
going to do? You're going to,well, I'm just going to keep
what I have, right? Or you haveto pay twice as much for that
next great thing. Well, okay,the rich people maybe they can
afford that, but there's not somany of them, and most people
will just hold on to what theyhave, or I'll sell you a really
(37:24):
crappy thing for half the price.
Okay, maybe some people will dothat too, so, but I feel it's
the fuel of a lot of markets,both B to B, B to C, where you
can come up with that next thingevery 18 months. That's so, so
wonderful, so fantastic. At thesame price that I got, I got to
throw that old thing away andget the new one right, but when
that slows down. So now, what doI mean so, so Moore's Law at
(37:48):
somewhere around, and this isdebated, but I would say
somewhere around. Let's say 20nanometers, the cost per gate
started going up all right? Thisis like gravity changing
directions, right? This iscompletely unheard of. So for
(38:08):
for since the 1950s until, let'ssay, five years or so ago, the
cost per transistor, or the costper gate, was going down every
generation by a lot, okay, andthat created all this magic that
I talked about somewhere aroundfive or so years ago. That
stopped, and the curve startedgoing up. So I could buy smaller
(38:30):
transistors at, say, 14nanometers, but those smaller
transistors at 14 nanometerscost me more than those
transistors did at 20nanometers. Now, why would I do
that? Well, okay, because I canget more performance. Okay,
fantastic. You have moreperformance, but you have more
performance at a higher cost.
(38:51):
Used to, used to be able to getmore performance at a lower
cost, right? That was the magicof Moore's law, and that's the
part of the, I would say,unbelievable revolution in the
market, in the industry, that'shappening now that I think is
completely underestimated. Ithink people just don't have a
clue how big of an impact. Okay,so, so what do you do? Yeah,
(39:13):
that was my next. Oh, my God,like the world's kind of doing
it. Well, the good news is, isthere's always an angle, right?
So I would say one thing thatMoore's law did, and maybe,
again, I'm being dramatic, butwhat it did is it made a lot of
sloppy, lazy engineers. So ifyou were given twice as many
(39:36):
transistors every 18 months toplay with for free, or the same
price that you paid before, ifyou're a silicon designer, you
say, Okay, your goal is to keepthis thing roughly the same die
size, so it's the same cost asyour last thing and But now,
instead of a milliontransistors, you have 2 million,
or instead of 5 million, youhave 10 million. Yeah, what do
you Okay, well, what do I do?
Well, okay, I had one ARM coreand so much memory. Now I'm. To
(39:58):
put in two ARM cores and a bunchof memory, right? Check, job
done. Right. Fantastic, right.
So now I got a processor that'smore memory and two ARM cores
versus one ARM core cost thesame as the last one. Boom. I'm
in Europe. I'm a rock star,right? You go to the customers,
Oh, I love it, right? And thenyou go from two to two to four,
(40:21):
and you go four to eight andmore and more memory. And so you
have ARM cores, very you know,simple instruction sets, lots
and lots of memory. So sloppycode is okay, fantastic, right?
(40:42):
So you get sloppy code on asimple instruction set and lots
of cores at the same price. Soeverything's working right.
Everything's fantastic. Keepdoing it right. Oh, wait a
second after 20 nanometers. Nowthat doesn't work anymore. Now,
okay, what do I do? Well, how doI get more performance? How do I
(41:02):
get lower power? How do I get soyou know what that is going to
create, or what it is creatingis more specialized processing.
So risk processors reducedinstruction set. Processors are,
by definition, you know, verylimited vocabulary, very
limited.
Chander Chawla (41:21):
So all arm is risk for people who are not from
the semi world. Yeah,
Mike Polacek (41:26):
and the instruction set of a processor
is, is I think of it. Maybe thisisn't the greatest analogy, but
it's, I think of it as, like thevocabulary of the chip. So if
you knew only, like 20 words inthe English language, you could
probably describe things, but itmight take you a long
Chander Chawla (41:42):
time. Yeah, right. You can describe limited
things. You can describe
Mike Polacek (41:46):
limited things and but you might use, you know,
those words over and over andover again. And, you know, if
you're talking about, you know,I don't know, you talk about a
pen, is a stick that writes withink. Okay, if you don't have the
vocabulary word for pen, but youhave those other words, it can
take you four or five words. Youcan use more energy to do it'll
(42:08):
take you longer, but so that'sthe instruction set. But if you
have a word for the job thatyou're doing, if you have an
instruction for the job you'redoing, if you have a complex
instruction set, that means youhave a lot more, a lot richer
vocabulary to get the job done.
And these are the jobs. What arethe jobs? The jobs are mostly
like math problems and thingslike that that you're doing at a
low level over and over and sowith a complex instruction set,
(42:32):
you could get more of the jobdone in a single cycle. And if
you do get more of the job donein a single cycle, then it's
more power efficient, and it canbe much higher performance at
the same sort of clock speed.
Yeah. So, and if you look at it,and when we were doing this at
(42:55):
Knowles in the voice processingspace, if you use sort of an ARM
processor to sort of brute forcedoing voice processing versus a
DSP that was made specificallyfor that job. Now that DSP isn't
going to be able to do lots ofother things that an arm
process, ARM processor, coulddo, but it could do that
specific job with thatinstruction set very well. And
you could, you could get sort of10x to 100x efficiency. Okay,
Chander Chawla (43:20):
how does that affect, you know, the size and
the power, if you get 10x moreefficiency, well,
Mike Polacek (43:27):
so you can do the same job at 110 of the power,
1/100 power. Power is just howmany transistors are wiggling
and how fast and so. So ifyou're doing something in one
cycle compared to something thattakes 10 cycles with the same
number of transistors wiggling.
That's just that capacitance isbeing charged or discharged,
right? So when
Chander Chawla (43:49):
you say powers, I immediately, oh, sorry,
performance. I immediatelythought, Okay, 10x more
performance that needs morepower requirement. Now it could
be the other way around thatsame thing. Yeah,
Mike Polacek (43:59):
yeah, yeah. So, so it's just a more it's more
efficient at doing that job andand so that's what you see
happening in a lot of marketstoday, is, is the movement
towards specialized processing?
Now, the industry isn't quiteready for that, because we have
generation, at least ageneration, maybe two
generations, of engineers thatwere living in this. Transistors
(44:21):
are free risk. Processors arethe bomb sloppy codes. Okay, you
know that that was, that was theworld that for decades. So now
we're in a world where, okay,now you got to actually create a
processor from scratch with yourown instruction set, like I got
to go to like, you know, a 1970stextbook to figure out what the
(44:44):
heck that means, you know, andand then you got to write very,
very tight code on that weirdinstruction set. Like I came
there's no, there's no, there'sno engineers that know how to do
that. Like, who. Who knows howto do that? So, so, you know,
it's good, it's going to beinteresting, you know, to see.
And so if you get, you know, agood example in today's world is
(45:07):
like tensor processor, right? Soat Google, and so they view a
TPU, right? So, you know, that'sa special processor specifically
to do the jobs that they want todo in the data center, right and
and so they created that theyhave very smart people there,
(45:30):
and they have enough of a scalethat they have enough of those
smart people that could do thehardware and could do the
software. And it all works in afairly closed environment, at
first inside Google, and then itsort of spreads out where, okay,
the hardware is still in theirdata centers. But although I
think now AWS probably hastensor their own, they created
(45:51):
their own. They have their own.
But I wouldn't be surprised thatthey have also tensor
Chander Chawla (45:58):
in their data
Mike Polacek (46:00):
center, in their AWS business, I would be
surprised they didn't. So thosestarts to move out, and then
people learn how, more and morepeople learn how to write to the
tensor processor, and then onand on so that that's created.
But then you're right. So now toAmazon, and these are sort of
the big, yeah, these are the bigexamples. Now there's like 1000
small examples where you haveindividual companies saying,
(46:22):
Well, okay, I want to just dothis job, and I but, but I need
that magic of Moore's law, and Idon't get it. So how do I reduce
the power dissipation of this,this device that I want to sell
into the market? Well, the onlyway I can do that is I, if I
have a custom processor. Well,and I don't, I don't know how to
make a custom process. I'm justlike a guy that makes, you know,
(46:43):
whatever I make, TWS, headsetsand, but I need a custom
processor, so I've got to gotalk to a semiconductor guy to
go do that. And, and so I thinkthat's a big transition that's
happening in the market thatwill probably play out over the
next 10 years, you know, and,and, you know, if Moore's law is
double every 18 months, a 10x to100x improvement is several
(47:04):
generations of Moore's law,right? Yeah. So let's call it
three to five or something, youknow. So, so let's say that's 10
years, but then at some pointthat's going to stop, yeah, at
some point you can't. You can't,sort of, you don't get that 10
to 100x improvement, like everygeneration. You get it once,
(47:25):
yeah, you get it once, goingfrom arm to not arm, right? And
maybe you can get a 10 or 20%improvement from there, but
you're not going to get thiseight, you know, 2x every 18
months thing anymore. So eventhough I think that there's sort
of hope here for the next 10years, let's say very
interesting, if you can find itnow. Who's going to win? I would
say the companies that have thebest hardware and software
(47:47):
engineers that can understandwhat's needed in terms of
workloads and and the special,specialized processing hardware
architecture and softwarearchitecture. Those are the
companies that are going to winthat see that and do it. But
then after that 10 years, Idon't know, I don't know, I
don't know what's gonna happen,then I think maybe there'll be
some, you know, there could besome breakthrough in the
(48:08):
hardware side. There could be,you know, quantum
Chander Chawla (48:11):
community, yeah, that could
Mike Polacek (48:13):
be like self driving cars too
Chander Chawla (48:17):
far. This is so fascinating. You know, I've
known you for 10 plus years, butevery time we talk about these
things, I'm like, Oh, wow, thisis fascinating. So I think we
are at time we should end here.
Want to say any parting wordsMike, as more advice to
entrepreneurs or people wholisten to us, who are mainly in
(48:40):
the Nordic countries, and theyare mainly in the software
world. But I think there's lotof things that we can learn from
the semi world, thinking aboutmaking sense of ambiguity,
taking risk, finding markets. SoI think any pardoning words in
that.
Mike Polacek (49:01):
I mean, I would say the only thing I could think
of that's worthwhile to end on,maybe, is, is that there's, you
know, these companies don't havea monopoly on innovation. They
don't have a monopoly on greatproducts. All the companies that
you see today, you think, Oh, myGod. Like, how could anybody do
anything bigger, better, orwhatever, than them? And I've
(49:23):
seen it. I'm old enough now, andI've seen it so many times in my
career, where, you know, soon asa soon as you say that, then
somebody comes out withsomething better, yeah, and and
so, and it doesn't have to begiant companies. It doesn't have
to be, you know, take over theworld at the beginning, but, you
know, small teams of people cando really great things and and
(49:44):
really, you know, changemarkets, change change, change
the market environment on theirown. So, you know, be confident
that as you're doing the workthat you're doing, be bold. Be
you know, take chances and. And,and there's nothing saying that
that the next successful startuphas to be out of Palo Alto. It
(50:07):
can. It can be anywhere in theworld,
Chander Chawla (50:08):
yes. Well, thank you very much, Mike for joining
us, and thanks to our listenersfor joining us. Arne will be
back next week, and I will beback in Palo Alto. Thanks. Thank
you. Bye.
Hello everyone.
(00:00):
Welcome back. This is ValleyNordic a podcast about culture,
(00:02):
technology and business from twoperspectives. That's my
perspective from Silicon Valleyand arne's perspective from the
Nordic countries. But today wehave a special guest, you'll get
two Silicon Valley perspectives.
(00:24):
I am with Mike politek InPescadero. That's the place
where Gordon Moore was born andraised. You. I'm sure I've heard
of Moore's law. So that GordonMoore was raised here, and Mike
politek is one of the smartestguys I know, and I had the
(00:45):
privilege of working with him,and I would say he I usually
like running my own show, butI'm happy to be his number two.
Anywhere you it's hard to findpeople who have the intellectual
depth and also who are ability,you know, their their ability to
communicate, you know, verycomplex idea in simple terms,
(01:07):
and align lot of people. So youwill see that today. So Mike,
welcome to Valley Nordic, andthanks for joining us. Thanks,
Mike Polacek (01:14):
Chander, that's introduction is too kind. And
also, I'm glad to do this faceto face, I don't know, for
breaking any rules, any laws.
Hopefully not we can edit thisout.
Chander Chawla (01:26):
We are both vaccinated, and we are in Mike's
beautiful place in Pescadero,and we're six feet apart. We are
six feet apart, and Mike has avery and I think everybody who
is a regular to valley Nordicknows my affinity towards
Berkshire, Hathaway to Buffettand Munger. So I see Mike is one
(01:49):
of the few executives who hasbeen on both sides, meaning he
ran big divisions. He waspresident at Knowles. Before
that, he was SVP at NationalSemiconductor, but so he has run
large businesses, created newbusinesses, and he spent four or
(02:09):
five years on the investingside. So he he's an operator and
also an investor. So which youknow, if, from the startup
perspective, the biggestchallenge you have is the
capital allocation. So you dothat internally, within a
company as a CEO or a startupfounder, but from externally,
(02:33):
you do you spread it apart toother companies, so you have
that perspective. So I find thatvery interesting, and today
we'll discuss, we'll do a deepdive into semiconductors. You
know what they are, why they'reimportant, and basically past,
present and future. And Mike,you want to say something before
(02:55):
I dive I'm so eager to dive intoit. But if there's anything you
want to add to the intro? No,
Mike Polacek (03:01):
I would just say we could definitely dive in that
your comment about investing andportfolio management, of your
investments and capitalallocation, resource allocation
within a company is more or lessthe same sort of idea as if
you're doing it outside, eitherinvesting in a division or
something you want to buy, ifyou're a strategic investor, or
(03:22):
if you're just a public marketinvestor, investing in stocks,
it's sort of the same thing thatthe diligence that you do in
terms of evaluating the productsand the technology and the
management team, and thenultimately valuation and all the
rest of it is, you know, it'sthe Same, same general concept.
And so I found that doing thatkind of work inside national
(03:45):
was, you know, quite applicableto doing it outside as
professional investor.
Chander Chawla (03:49):
Yeah, great. So I want to start with the
fundamentals of siliconbusiness. Because, you know, a
lot of the training people getin the startup world. When these
days, when people think ofSilicon Valley, they think of
basically B to B, SaaS startupsor consumer companies like
Facebook, Google or slack wouldbe B to B, example. But it
(04:15):
started, it's called SiliconValley because of
semiconductors. So how are thein your view, the fundamentals
of silicon business differentthan, let's say, A, B to B, SAS
startup.
Mike Polacek (04:29):
Well, I think the most obvious first thing I can
think of is just the investmentrequired and the time frame
required to get a product out.
And you know, it's verydifficult to sort of spend a
little bit of money testsomething, see if it works,
adjust, modify, get somethingelse out incrementally, a little
(04:49):
bit better. You know, that'sjust not how semiconductors
work. Generally speaking, youmake a fairly large investment
depends on the complexity of thechip and sort of. What you're
working on, but, but you canmake a multi year investment, a
multi per multi 100 person yearinvestment, 10s of millions of
dollars investment, before youcan try the product out and and
(05:12):
that that requires, I would say,a lot more discipline. There's a
lot more risk, but then you needa lot more discipline up front
and and then you need to, youneed to hit big, big markets,
because those investments arelarge and the risk is large. So
so then you better be, you know,fairly sure that you're going to
(05:34):
be able to address a market thathas the right number of zeros
for that to have been a goodinvestment,
Chander Chawla (05:42):
and how do you know that that market is there?
Mike Polacek (05:47):
Well, you don't know. I mean, I would say you
could invest in a market thatyou're already in. So if you're
already in a market and you'resort of doing incremental
improvements, then the risk islower. You you may have gotten
this feedback from yourcustomer, or your customers
customer, that this next thingthat you should do, that there's
(06:07):
a demand for that, even evensaying that, you know, even if a
customer comes to you, whichhappens all the time, a big
customer will come to you andsay, oh, I want this next
version of this last thing youhave. But guess what the that
customer is saying the samething to your competitors, yeah.
So you're racing to try to getthat product out, to sort of
(06:27):
meet the performancerequirements, get the cost
structure in place so you canhit price targets, get the time
frame right, and you do all thatinvestment for that customer
that says, I want this or a setof customers. And then you get
there and you deliver samples,and you find out that, well,
your competitor did it a littlebit better, or a little bit
cheaper, or whatever, and nowyou're sort of in the back seat
(06:51):
at best. So there's still riskthere. The other opportunity is
to go after a market that's new,which is riskier in some ways,
in terms of, you know, is thatmarket really going to be there
when you arrive two years, threeyears later, at the size that
you need it to be? And then, butthen the risk goes down,
necessarily, or maybe notnecessarily, but the risk
(07:12):
certainly goes down with respectto competition. So it's not as
obvious a market. Maybe, maybeyou've seen trends that other
that you've maybe yourcompetitors haven't seen. And so
then, when you arrive at thatmarket, and that market's there,
there may not be as manycompetitors.
Chander Chawla (07:29):
It's, I remember you told me this, and I still
remember this, because it madean impression. You know, LED
lights, the first commercialavailability was in like, 65
Mike Polacek (07:41):
Well, the, yeah, the invention, you know, it was
a University of Illinoisprofessor that my alma mater,
and it was, it was decadesbefore I was there, and I'm not
young, that he invented LEDs.
And he actually made aprediction about that someday
that LEDs would be used forlighting instead of incandescent
lights. And, yeah, it took manyyears, you know. And that's, you
(08:04):
know, I always say that thebigger challenge in identifying
new markets, at least for me, myexperience, has been, for my
career, has been the timing. Youknow, more than Are you right or
not right? You know, I wouldsay, yeah, you can look at LEDs
in the 1950s or 1960s and say,Yeah, this is a great technology
(08:25):
that emits light, and it'sinherently lower power, and it's
but okay, there's a bunch ofthings that have to fall into
place for them to becommercially available. So when,
if you're sitting there in 1950or 1960 you know, write down
what's the year you think LEDswill become popular, you'll be
able to buy them at whateverhardware store. And, you know,
being able to do that accuratelyis very difficult, yeah, I
(08:48):
remember, you know, in selfdriving cars. So this was, you
know, I think it was maybe 567,years ago, when they were first
starting to kind of get into thedialog of, of, okay, we might
have self driving cars and andthere's different versions and
different levels, and maybe, youknow, trucks and busses and all
(09:11):
these so. So this was, like abig thing in 2000 I don't know,
1415, something like that. And Iremember reading or listening to
and I hope I get this right ofthe seats. I think there's a CEO
Ford, or some high levelexecutive at Ford coming out and
saying, you know, by in the nextfive years, this was in 2015 or
(09:33):
2016 so basically, by today,2021 Ford will be making cars
without acceleration,accelerators, pedals, steering
wheels. That was going to bemainstream, mainstream enough
that Ford's good, right? Thisis, this is, you know, and I
remember reading that thinking,here's a guy that just doesn't
(09:56):
understand the technology,right? But I'm sure he's very
smart, very accomplished, but hegot that wrong. Now, it doesn't
mean that that's never going tohappen, that that's an absurd
idea. It's like a crazy he's notbreaking laws of physics by
saying that he's he's justreally wrong with respect to the
timing, and that's, I would say,really wrong related to the
(10:18):
complexity of the problem? Yeah,and I think you just
underestimated the complexity ofthe problem. And I think you see
that kind of thing in a lot ofplaces, in a lot of markets,
where it's sort of obvious, thisis going to happen, but where
you really need to be right isin the level of detail about how
is it going to happen, when isit going to happen? Where is
(10:39):
going to happen if you're reallygoing to exploit a big trend
that's maybe fairly obvious, youknow, you need to dig several
layers down. And I would saythat's a much bigger challenge.
Chander Chawla (10:50):
Is there a method or system people can
follow to do that?
Mike Polacek (10:56):
I don't know. I I think there's different phases
of trying to identify newmarkets. So, so one might be
like a brainstorming phase. Sowhat's possible? You know,
linear thinking is problematic,I would say, to being creative.
Chander Chawla (11:15):
Say more that's interesting, like, say more on
why linear thinking isproblematic.
Mike Polacek (11:20):
So I mean linear thinking will get you linear
results, so it'll get you thenext thing like and that may be
great. There's nothing wrongwith doing the best next thing,
and that's an aspect ofinnovation, for sure. But if you
want to do something a littlebit bigger impact than that,
then generally those things arenonlinear, meaning they don't
(11:41):
follow that. Okay, six monthsago, this improved this much, so
six months from now, this isgoing to improve this much,
like, that's linear thinking.
And most and humans, I wouldsay, in general, are pretty
linear thinkers. Like, it's justsort of, I don't know it's in
our DNA or hardware wired, likewe don't react well to, you
know, exponential curves, like,we sort of can't get our heads
(12:03):
around that. And you see thateverywhere in society, people
just don't, sort of can't makethe connection between, what's
the difference between a billionand a trillion, or what's the,
you know, the one of these gamesI remember is, is, would you
rather have a million dollars.
I'll give you two options. I'llgive you a million dollars
(12:24):
today, or I'll give you a pennytoday and I'll double it every
day for 30 days, right? Turnsout, if you take a penny today
and you double it every day, onepenny one day, two pennies the
next day, on and on that thatsecond option would be a much
bigger number than the firsttime, right? It's just hard to
think like, yeah. You know, it'sjust people don't think that
(12:47):
way, yeah. So, so in order to bea non linear thinker, you have
to get out of that trap. And howdo you get out of that trap? I
think there's certain, you know,there's certain brainstorming
exercises you could do that are,you know, to try to try to get
yourself, you know, one is, isthis? Who's listening to this
(13:09):
guy? I think his name is deBono. Edward de Bono, if I have
that right, he's, he's got thesesix hats kind of thing. And Sri
has used this, I think Sri hasused this methodology. They have
versions of this where they yousort of play different roles.
You get a group of peoplebrainstorming, and people are
sort of play different roles toattack and to, you know, to and
you sort of get out of, try toplay, you literally play roles
(13:32):
to try to get out of theconventional thinking. You know,
there's, there's, there's thingsof looking for convergence,
right? Looking for two marketsthat exist and that are
happening and that you're sureare going to happen. And then
you just sort of write themdown, put them next to each
other, and say, it can be stupidstuff, right? It could, but, but
(13:52):
there could be some, some goodthings, right? So you take, you
take 5g which is like a bigobvious thing, and you take IoT,
which is a big obvious thing,and then you stick them together
and and then you sort ofbrainstorm what could happen if
you bring 5g kind ofconnectivity to a Iot, type of
(14:13):
sensor network world, right?
Okay, there could be all kindsof cool things can happen. You
could imagine and on and on andon. You could look for gaps, you
know, in in technology, whereyou sort of hit a wall and then
something, you know, what couldsort of extend that wall. So,
yeah, you could look for, youknow, all kinds of different
(14:37):
ways to kind of get yourself inthis creative and non linear
sort of thinking world. Yeah,
Chander Chawla (14:43):
it's hard, because you're right. Most
people think that way. And youthink about what's taught in the
software world, in the MVP, andhow to take products to market,
that's what's taught. You know,you create an MVP that's.
Minimum viable product, and youget it out in the market as fast
(15:04):
you can. And then you iterate,yeah, you know, get feedback,
which makes sense, it becauseit's software, but like you
shared in the very beginning,the fundamentals of the
semiconductor business aredifferent. You can't really
iterate your way to success. Imean, unless you have billions
of dollars, you can iterate, butas a startup, it's hard to
(15:26):
iterate your way to success withlinear thinking or incremental
improvement and capture anythingbig. Yeah,
Mike Polacek (15:35):
and, you know, I mean lots. There's, again,
nothing wrong with incrementalthinking and software or
anything else. I mean, a lot ofthe I think people also
overestimate, sort of like anindividual God, like creator
that invented, magically, theiPhone or the or the PC or the
(15:56):
operating system like the like,if you know the details of
these, Every one of thesestories there, it's, there's
nothing sexy about like, it's,it's 1000 small things come
together to create a greatproduct. And, and it's maybe
that attention to detail ofbringing them together, and the
attention to detail of executingon a schedule. And, you know,
(16:17):
those kinds of things which arenot sexy. You know, read them in
the big books by,
Chander Chawla (16:20):
you know, Walter Isaac by Walter Isaacson,
Mike Polacek (16:23):
exactly, so, but, but that's usually, you know,
the story behind the iPhone, orthe story behind, you know,
Microsoft OS, or something like,it's not, you know, there's,
there's not, usually, now,sometimes there is magic, but
Oftentimes the magic isaccidental, yeah, which? Which
is also this nonlinear sort ofthing. A lot of like, the
(16:45):
coolest inventions areliterally, like, lab accidents
and things like that. Where,yeah, that's, how do you
capture? How do you make sureyou're ready for the accident
when it happens? You know,that's also a mindset. If I see
something that's a surprise thathappens, what's my reaction? Is
my reaction, oh, let's throwthat in the garbage because it's
a surprise. Or is my reaction,oh, my God. Like, this is really
(17:08):
cool. Like this, this is, thiswasn't supposed to be this way.
Let's, let's, let's, let's,let's do some more work on that.
Chander Chawla (17:13):
Yeah, yeah.
It's, can we do two minutessidebar on the accidents? You're
right. Lot of them are accidentswith things moving to remote
work and hybrid work. What doyou think happens to those
accidents?
Mike Polacek (17:32):
Well, some, some can be done by individuals, I
suppose, like a lab, a lab.
What's a lab in the SAS world?
It's maybe somebody's laptop intheir kitchen, right? I mean, I
I'd have to think about it somemore, but, but, I mean, I would
say, certainly one of the thingsyou miss is the serendipity of
(17:53):
human brainstorming, person toperson interaction, where you
have an idea and you bump intosomebody. I bump into you. I
said, Hey, I was just thinkingabout this. I had this weird
dream, and I wrote down thesenotes at two in the morning, and
I and I woke up this morning andit's like, oh my god, this is a
great idea. What do you thinklike? But is it crazy? And then
I go, you, and I say, Oh, isthat crazy? And you say, well,
(18:14):
it's a little crazy, but let'sask Jim. He's right over here.
He's an expert on this,whatever, blah, blah, blah. So
that kind of thing happens,right? That kind of thing
happens a lot, where you cansort of iterate, brainstorm,
bounce things off each other.
You know, the barrier that's notimpossible to do in a sort of
virtual environment, but thebarrier to do that is higher,
(18:34):
like the am I going to call youor I'm going to schedule a zoom
call with you, because I hadthis weird dream. I would say,
No, it's a little weird. Iprobably wouldn't do that,
right, but I might just mentionit to you by the coffee machine.
Yeah.
Chander Chawla (18:50):
I mean, generally, I found if whenever
we had these conversations orwith other people, they're more
in casual environments, we arehaving coffee, or we are having
a beer, or like, becausenormally during the work day,
you're so focused on what youneed to deliver the quarterly
(19:11):
results, or whatever it is, thisis more when you're in a
different mind frame and that Ifind it hard. You know how we'll
see how it get it getsintegrated in the remote work
world,
Mike Polacek (19:22):
yeah? I mean, it's certainly more fun to do this
kind of thing face to face, youknow, versus, you know, sitting
around by yourself, working ontasks all day, yeah? Which is,
you know,
Chander Chawla (19:35):
okay, let's come back to semiconductors. Can you
like, I always found thisfascinating, the value
distribution, you know, atnational we were making these
chips, which were, you know,like, basically, you couldn't,
let's say, ship a phone withouta PMU. But PMU is power
(19:58):
management unit that. Helps youknow how the power to distribute
it to different parts in thephone. So if that was, maybe we
can share it like sub $2 parts,and the actual bomb or bill of
materials of the phone was atthat time, let's say, 200 bucks.
(20:20):
So that's like, 1% and like Ifelt everybody we are doing, you
know, let's go back to the firstthing you said, how much risk it
takes, how much investment ittakes to get things right. So
the semi people are taking allthis risk, failing multiple
times, to get these thingsright, and making investments,
(20:43):
and then it goes to the phonemanufacturer, which makes huge
margins, lot of money. Then theoperator, who sells the phone or
has a service plan, they need alot of money. So the value
that's created in thisecosystem, how is it that the
guys were doing, or gals, thetaking the most risk, are
(21:04):
getting least reward.
Mike Polacek (21:07):
Well, I would say the margins on those devices
were high, probably higher thanthe phone guys were making. So
that's one indicator that thatthere is value being placed on
it. So maybe not value in termsof margin dollars per phone, but
in terms of margin percent, itwas high. I think the risk part
(21:28):
of it is, you know, it's aconscious decision. If you're
going to go after a socket in aphone, it's you're making a
conscious decision that I'mgoing to go after a market where
the power dynamic because of thevolumes of the phones and the
concentration of phone makersversus the concentration of
suppliers. That I'm making aconscious decision, that I'm I'm
(21:52):
taking on a risk there. I'mwilling to, I'm going to be
playing a subservient role wherethis guy is good, this this
phone guy is going to push me tobe able to push me around if he
wants to. And, but why do that?
Well, you do that because,because that Phone Guy might be
shipping 100 million units ayear. You know, tell what other
market could you? Could you, youknow, for a couple bucks, could
you get into a market thatyou're that that one guy could
(22:13):
give you a design win that's,that's 100 million units. You
know, it just doesn't exist. Youknow, you get excited about a
design win in a in a car or a TVor whatever, you'd be lucky to
ship a million units, right? So,so now, my personal opinion is
that that's a that's date like,it's very dangerous. It's a very
potentially is a big, bigmistake. And there's a lot of
(22:36):
companies around, startupsaround, that no longer exist
because they were taken out bythese companies. That these
companies are prettybloodthirsty, you know, they'll,
they'll sell your theirgrandmother for a nickel. And so
if you, if a competitor comesalong, and oftentimes, these
phone guys will, will guide yourcompetitor, if they feel like
(22:58):
you're you have a little toomuch power though, maybe somehow
the planets aligned and you yougot something that no one else
could do. And maybe you want tocharge a little more. You want
to do whatever the Phone Guysays, not in my market. So
they'll help their competitorand then literally take them out
of business. And that's, youknow, if that's your only
(23:20):
customer, you get very excited.
You get 100 million unit deal.
You go from zero to $200 millionin revenue. All of a sudden,
champagne corks, maybe you gocrazy, and you want to go
public, you do all these kind ofthings. And then, and then, all
of a sudden, that that's acustomer says, oh, you know, I
really don't like that productanymore. You should cut the
(23:41):
price in half, or I'm gonna usesomething else. And you're like,
what? And then you implode, andyou have to fire everybody, and
you cry. You don't have yourselfto sleep. So So I would say,
Yeah, you should go into thatwith your eyes open. And there's
certainly other markets whichare smaller volume where that
power dynamic isn't the same. Infact, the same. In fact, the
power dynamic may be theopposite. The power dynamic may
(24:03):
be that you know you're youbecome such an important partner
to that customer that they treatyou with respect and longevity
and loyalty and all these. Andyou could do it mutually. And so
now you won't ship 100 millionunits in a year, but it's sort
of, you know, what kind oftrajectory Are you looking for,
and how stable a business areyou looking for? So these are
(24:25):
conscious decisions. I would sayneither is right or wrong. If
you're in a huge company likenational or some other company
where we had 1000s, literally1000s of products, you know,
maybe a dozen, couple dozendifferent product lines. Yeah,
you can have a portfolioapproach where some of them,
some of my investments, aretowards this very high risk, but
(24:47):
potentially home run typevolume, and then other
investments are on the otherextreme, and lots of between.
And I would say that would be asmart way to deploy your assets.
Chander Chawla (24:59):
Yeah. It's interesting. You know how that
happens? You think the cycle,the phone replacement cycle, has
something to do with it?
Because, you know, peoplegenerally replace their
smartphones every two years, orand companies release their
phone every year, so they have achance to change things every
(25:19):
year. I remember when I workedwith Siemens, you know, you were
selling network equipment, sorouters and big switches and
DWDM stuff, and they it's intheir network. If something
fails, they lose business. Sothey always wanted three vendors
(25:39):
for everything they did. So ifsomething fails, but the
relationship was there, theycouldn't just throw you out,
like in the phone business, theycan just throw you out next year
here that you're integral to howthey make money here, that cycle
changes every year. Do you thinkthat has something to do with
it? Because the fast cycle?
(26:01):
Well,
Mike Polacek (26:02):
I mean, the consumer market definitely has
less of a requirement forloyalty than other markets that
are longer cycle markets. So Ithink that's right, you know,
will it? I mean, you know,replacement cycles are
lengthening in phones, right? Itused to be 20 months, and then
it was 22 then it was 2426 28 Ithink it's probably getting
(26:22):
close to two and a half yearsnow, instead of two years. And
so, you know, will that breedmore loyalty? I doubt it. You
know, it's, it's not somethingthat you know. I think it's, I
think it's the volume you know,more than the, than the cycles
(26:45):
where, where the, just thevolume is so high. And every
penny, half Penny, you know, Isaw, I saw this. I mean, it's
some at pretty much all thesecompanies that I've worked with
that were good customers, thatsort of became not so good
customers, you know, we saw itwith IBM in the literally 80s
and 90s, saw it with Nokia,where we used to have meetings
(27:07):
with the design engineers, likewe would be semiconductor guys,
we'd have all this greattechnology we were thinking
about working on, and then we'dhave the design guys for the PC,
and they were talking about whatthey needed, if they could dream
of something, what would it be?
And we would write it down, we'dcome back a month later, and we
would have those kind ofconversations, and and then the
(27:27):
planets would align, and then atsome point, you would come and
you would have developedsomething, and you'd give them
samples, and they would use it.
And everybody loved each other,right? And, and they got a
great, innovative solution outof it. They got something then
that they could do somethingthat their competitors couldn't
do. And, you know, we did thesame thing. Yeah.
Chander Chawla (27:54):
So the I think Mike has spent a lot of time in
the industry I spent limitedtime so you're seeing how now
he's talking about how thingshave evolved, how it used to be,
and then what it became afterthe customer's success or market
position changed, yeah,
Mike Polacek (28:14):
the volume got bigger. So then what I saw is,
then slowly, what happened isthat the IBM every meeting, then
at some point, the procurementguys would be there. And so then
it would be like in the in sortof product engineering guys. So
then you would have like, halfthe people in the room were
design and design guys, and thenthe other half were sort of
more, sort of plumbing guys. Andthen a year later, and then
(28:37):
there'd be like one design guy,and he wasn't allowed to ask
questions, you know, and it wasall the procurement and product
guys and then, and then, in theend, like you couldn't, you
know, you're later, like youcouldn't meet the zeitgeist like
the other guys, they just wantedto talk about your existing
products and what you're doingto cost reduce them, and what
you're doing. And maybe theproduct guy would say, you know,
I need something with slightlylower power dissipation or
(28:58):
slightly higher performance,whatever. I'm being dramatic.
But it was sort of that story.
And that same story happenedwith Nokia. I remember Nokia, we
used to have these greatconversations with Nokia, these
design engineers like, fly allover the place, and we would go
there, and they would come hereand and there'd be all this
designer to designer type, greatstuff happening, technical, deep
technical conversation aboutwhat's possible, like, and
that's the magical interfacebetween, like, what's possible
(29:21):
and what's needed, yeah, right.
And so you both togetherdiscover that, and then we would
do these investments and do thiswork. And it wouldn't be
foolproof, but it would, yeah,you would create really cool
stuff out of that. And, but, butthen Nokia became this giant
phone company, and thosemeetings changed. You know that,
(29:43):
you know, and I think, fromtheir side, why do they change?
And their side, they're like,well, we don't want any like
silicon supplier to get too muchpower. So we don't want you,
like, in the back rooms with ourtechnical people to sort of
position me the procurement. Guyinto a sole source position
where I really want somethingyou have, and you're the only
(30:04):
guy that can supply it. Sothey're trying to prevent that
from happening where, where youhave this really important,
unique product that they need,right? And it's sort of built
into the system. And then I sawit, you know, again, you know,
more recently with companieslike Apple, that is sort of the
same thing. I remember apple. Weused to have these great
(30:25):
conversations with Appledesigners, and now I don't
million years. You don't get toget the same building to ask
people
Chander Chawla (30:34):
that, I mean logically. Well, it depends what
logic you look at. So what youjust described the procurement
team or bigger companies aredeliberately saying, we don't
want to innovate, becauseinnovation in the ecosystem
comes from one initially, fromone company. So one company is
(30:55):
going to have a breakthrough,and you can have access to that.
But what, when the companies getbigger, it becomes, you know,
risk avoidance, rather thancoming to a new thing, you can
get first and have more revenue.
Is that? Do I get that? Yeah,
Mike Polacek (31:13):
I think it's, what do you emphasize? What's the
most important thing to you? Andat some point, the volume gets
so big that what becomes mostimportant to you, and this may
be totally correct, like forshareholder value or the look at
Apple. Apple's obviously beenunbelievably successful under
Tim Cook, but the emphasis hasbeen more on production,
(31:34):
reliable manufacturing, gettingmargins stable, being able to
ship in unbelievable volumes atreliably high quality. All those
kinds of things have become moreimportant than innovation. I
Chander Chawla (31:50):
would say, I would say, everything you said
is correct, but innovation,they're doing internally.
They're doing verticalintegration. So they're making
their own chips. They're, youknow, influencing Corning. So
they there's lot of innovationstill happening, but it all in
house. So they've reduced thedependency. They still want to
(32:13):
innovate. And they may havenoticed what you just described,
procurement people, or whateverreason you can't really get
innovation in, in, in the doorfast enough. So let's do it
internally, in,
Mike Polacek (32:25):
yeah, yeah. And I'm sure there's a lot of great
innovation happening there, Ithink, but, but the The old
saying is true still, that mostof the great engineers in the
world don't work for fill in theblank company. Yeah. If you're
sitting at Apple, you have to behumble and say, you know, most
of the great engineers in theworld don't work at Apple. You
(32:46):
know, the vast majority don'twork at Apple. No, that doesn't
mean Apple doesn't have greatengineers. Of course they do.
But if you can put two companiesside by side, one company that's
doing everything internally, andone company that's able to sort
of create a nurturing ecosystem,a bunch of people that are doing
that, they're doing work forinternally, but also doing work
for them externally. I couldtell you which company will be
(33:09):
successful long term, and, and,or at least, ultimately
successful, or most successful,
Chander Chawla (33:16):
it's like, let's take the Samsung example. You
think they're doing internal,external combo, or like they
compete with Apple, but Apple, Iwould say, still is, you know, I
Mike Polacek (33:29):
mean, there's been a huge trend. I mean, for a long
time during my career, there wasa trend all these system
companies were all putting theirsemiconductor stuff out, like
we're spinning them out, right?
Every single big company,Phillips, Siemens, IBM, right,
they were all spinning out. Why?
Because the capital investmentswere so large and and they were
(33:51):
getting larger every generation.
The factories were costing more,the development costs more. And
it was just like, Okay, we don'thave the volume to so we're
going to spit this thing out.
And that was going on fordecades, and definitely in the
last whatever five, 510, years,it's definitely been moving back
in, where you get more verticalintegration. Samsung's been
(34:13):
doing it. Huawei has been doingit. Apple does it. Even guys
like Google and Amazon you don'tthink of as hardware companies,
you know, have their own hugechip development, yeah,
organizations now. So, you know,why is that? You know? I would
say there's a few things, youknow, one which we could talk
(34:35):
about, if you want, now orlater, but sort of, what's the
impact of the of the slowdown orelimination of Moore's law?
Chander Chawla (34:45):
Yeah. Let's talk about it. Yeah. I think very
interesting, yeah. I mean,Gordon Moore's, you know,
territory. Let's talk about it.
Yeah, it's,
Mike Polacek (34:55):
you know, first, when I say Moore's law, what I
mean it so I think people talkabout it when you. Read things,
it's sort of, I think it gets alittle bit muddy, but what, what
you see is, can you still bruteforce shrink transistors? If
that's what you think of Moore'slaw, then, yeah, I think you can
still shrink transistors at somepoint. There's an end there. But
(35:16):
they're they're gone below 10,and they're going to seven, and
then people talk about three,and where's the wall, and you
get x ray lithography and gammaray, I don't know, all the sort
of you can make light smallerand smaller and and, you know,
so that part, I would say, Okay,I'm not even gonna argue that
maybe you can still shrinktransistors. Fantastic. But the
magic of Moore's Law was theeconomic side, was that you
(35:41):
could shrink transistors andmake transistors or gates
cheaper by a lot everygeneration.
Chander Chawla (35:49):
Yeah, that's around 18 months to two year
yeah, and,
Mike Polacek (35:54):
and, and that's the magic of a lot of markets.
So if you're a system company,and you know, every 18 months, I
can get something that's goingto vastly outperform the last
thing at the same price, thenyou can make a system around
that that's going to do the samething, and then you're going to
create demand for that newproduct, for that reason,
(36:16):
because now your new systemproduct is going to do some
wonderful things it didn't DoBefore at the same price, right?
Or maybe, if the other side is,you can do it at, you know, the
same performance, you can do ita lot cheaper, yeah, which is
another sort of dimension, yeah,and, and that's the magic of
Moore's law. And I would say, Idon't maybe this is hyperbolic,
because I'm a semiconductor guy,but I think that is the biggest
(36:40):
economic impact for the UnitedStates in my lifetime, that that
so many industries are sustainedby that concept of having the
next thing be so much better atthe same price every 18 Months.
Think about a world where thatdoesn't happen, where, where,
(37:02):
where your your choice, 18months later is to buy something
that's more or less the same atthe same price. What are you
going to do? You're going to,well, I'm just going to keep
what I have, right? Or you haveto pay twice as much for that
next great thing. Well, okay,the rich people maybe they can
afford that, but there's not somany of them, and most people
will just hold on to what theyhave, or I'll sell you a really
(37:24):
crappy thing for half the price.
Okay, maybe some people will dothat too, so, but I feel it's
the fuel of a lot of markets,both B to B, B to C, where you
can come up with that next thingevery 18 months. That's so, so
wonderful, so fantastic. At thesame price that I got, I got to
throw that old thing away andget the new one right, but when
that slows down. So now, what doI mean so, so Moore's Law at
(37:48):
somewhere around, and this isdebated, but I would say
somewhere around. Let's say 20nanometers, the cost per gate
started going up all right? Thisis like gravity changing
directions, right? This iscompletely unheard of. So for
(38:08):
for since the 1950s until, let'ssay, five years or so ago, the
cost per transistor, or the costper gate, was going down every
generation by a lot, okay, andthat created all this magic that
I talked about somewhere aroundfive or so years ago. That
stopped, and the curve startedgoing up. So I could buy smaller
(38:30):
transistors at, say, 14nanometers, but those smaller
transistors at 14 nanometerscost me more than those
transistors did at 20nanometers. Now, why would I do
that? Well, okay, because I canget more performance. Okay,
fantastic. You have moreperformance, but you have more
performance at a higher cost.
(38:51):
Used to, used to be able to getmore performance at a lower
cost, right? That was the magicof Moore's law, and that's the
part of the, I would say,unbelievable revolution in the
market, in the industry, that'shappening now that I think is
completely underestimated. Ithink people just don't have a
clue how big of an impact. Okay,so, so what do you do? Yeah,
(39:13):
that was my next. Oh, my God,like the world's kind of doing
it. Well, the good news is, isthere's always an angle, right?
So I would say one thing thatMoore's law did, and maybe,
again, I'm being dramatic, butwhat it did is it made a lot of
sloppy, lazy engineers. So ifyou were given twice as many
(39:36):
transistors every 18 months toplay with for free, or the same
price that you paid before, ifyou're a silicon designer, you
say, Okay, your goal is to keepthis thing roughly the same die
size, so it's the same cost asyour last thing and But now,
instead of a milliontransistors, you have 2 million,
or instead of 5 million, youhave 10 million. Yeah, what do
you Okay, well, what do I do?
Well, okay, I had one ARM coreand so much memory. Now I'm. To
(39:58):
put in two ARM cores and a bunchof memory, right? Check, job
done. Right. Fantastic, right.
So now I got a processor that'smore memory and two ARM cores
versus one ARM core cost thesame as the last one. Boom. I'm
in Europe. I'm a rock star,right? You go to the customers,
Oh, I love it, right? And thenyou go from two to two to four,
(40:21):
and you go four to eight andmore and more memory. And so you
have ARM cores, very you know,simple instruction sets, lots
and lots of memory. So sloppycode is okay, fantastic, right?
(40:42):
So you get sloppy code on asimple instruction set and lots
of cores at the same price. Soeverything's working right.
Everything's fantastic. Keepdoing it right. Oh, wait a
second after 20 nanometers. Nowthat doesn't work anymore. Now,
okay, what do I do? Well, how doI get more performance? How do I
(41:02):
get lower power? How do I get soyou know what that is going to
create, or what it is creatingis more specialized processing.
So risk processors reducedinstruction set. Processors are,
by definition, you know, verylimited vocabulary, very
limited.
Chander Chawla (41:21):
So all arm is risk for people who are not from
the semi world. Yeah,
Mike Polacek (41:26):
and the instruction set of a processor
is, is I think of it. Maybe thisisn't the greatest analogy, but
it's, I think of it as, like thevocabulary of the chip. So if
you knew only, like 20 words inthe English language, you could
probably describe things, but itmight take you a long
Chander Chawla (41:42):
time. Yeah, right. You can describe limited
things. You can describe
Mike Polacek (41:46):
limited things and but you might use, you know,
those words over and over andover again. And, you know, if
you're talking about, you know,I don't know, you talk about a
pen, is a stick that writes withink. Okay, if you don't have the
vocabulary word for pen, but youhave those other words, it can
take you four or five words. Youcan use more energy to do it'll
(42:08):
take you longer, but so that'sthe instruction set. But if you
have a word for the job thatyou're doing, if you have an
instruction for the job you'redoing, if you have a complex
instruction set, that means youhave a lot more, a lot richer
vocabulary to get the job done.
And these are the jobs. What arethe jobs? The jobs are mostly
like math problems and thingslike that that you're doing at a
low level over and over and sowith a complex instruction set,
(42:32):
you could get more of the jobdone in a single cycle. And if
you do get more of the job donein a single cycle, then it's
more power efficient, and it canbe much higher performance at
the same sort of clock speed.
Yeah. So, and if you look at it,and when we were doing this at
(42:55):
Knowles in the voice processingspace, if you use sort of an ARM
processor to sort of brute forcedoing voice processing versus a
DSP that was made specificallyfor that job. Now that DSP isn't
going to be able to do lots ofother things that an arm
process, ARM processor, coulddo, but it could do that
specific job with thatinstruction set very well. And
you could, you could get sort of10x to 100x efficiency. Okay,
Chander Chawla (43:20):
how does that affect, you know, the size and
the power, if you get 10x moreefficiency, well,
Mike Polacek (43:27):
so you can do the same job at 110 of the power,
1/100 power. Power is just howmany transistors are wiggling
and how fast and so. So ifyou're doing something in one
cycle compared to something thattakes 10 cycles with the same
number of transistors wiggling.
That's just that capacitance isbeing charged or discharged,
right? So when
Chander Chawla (43:49):
you say powers, I immediately, oh, sorry,
performance. I immediatelythought, Okay, 10x more
performance that needs morepower requirement. Now it could
be the other way around thatsame thing. Yeah,
Mike Polacek (43:59):
yeah, yeah. So, so it's just a more it's more
efficient at doing that job andand so that's what you see
happening in a lot of marketstoday, is, is the movement
towards specialized processing?
Now, the industry isn't quiteready for that, because we have
generation, at least ageneration, maybe two
generations, of engineers thatwere living in this. Transistors
(44:21):
are free risk. Processors arethe bomb sloppy codes. Okay, you
know that that was, that was theworld that for decades. So now
we're in a world where, okay,now you got to actually create a
processor from scratch with yourown instruction set, like I got
to go to like, you know, a 1970stextbook to figure out what the
(44:44):
heck that means, you know, andand then you got to write very,
very tight code on that weirdinstruction set. Like I came
there's no, there's no, there'sno engineers that know how to do
that. Like, who. Who knows howto do that? So, so, you know,
it's good, it's going to beinteresting, you know, to see.
And so if you get, you know, agood example in today's world is
(45:07):
like tensor processor, right? Soat Google, and so they view a
TPU, right? So, you know, that'sa special processor specifically
to do the jobs that they want todo in the data center, right and
and so they created that theyhave very smart people there,
(45:30):
and they have enough of a scalethat they have enough of those
smart people that could do thehardware and could do the
software. And it all works in afairly closed environment, at
first inside Google, and then itsort of spreads out where, okay,
the hardware is still in theirdata centers. But although I
think now AWS probably hastensor their own, they created
(45:51):
their own. They have their own.
But I wouldn't be surprised thatthey have also tensor
Chander Chawla (45:58):
in their data
Mike Polacek (46:00):
center, in their AWS business, I would be
surprised they didn't. So thosestarts to move out, and then
people learn how, more and morepeople learn how to write to the
tensor processor, and then onand on so that that's created.
But then you're right. So now toAmazon, and these are sort of
the big, yeah, these are the bigexamples. Now there's like 1000
small examples where you haveindividual companies saying,
(46:22):
Well, okay, I want to just dothis job, and I but, but I need
that magic of Moore's law, and Idon't get it. So how do I reduce
the power dissipation of this,this device that I want to sell
into the market? Well, the onlyway I can do that is I, if I
have a custom processor. Well,and I don't, I don't know how to
make a custom process. I'm justlike a guy that makes, you know,
(46:43):
whatever I make, TWS, headsetsand, but I need a custom
processor, so I've got to gotalk to a semiconductor guy to
go do that. And, and so I thinkthat's a big transition that's
happening in the market thatwill probably play out over the
next 10 years, you know, and,and, you know, if Moore's law is
double every 18 months, a 10x to100x improvement is several
(47:04):
generations of Moore's law,right? Yeah. So let's call it
three to five or something, youknow. So, so let's say that's 10
years, but then at some pointthat's going to stop, yeah, at
some point you can't. You can't,sort of, you don't get that 10
to 100x improvement, like everygeneration. You get it once,
(47:25):
yeah, you get it once, goingfrom arm to not arm, right? And
maybe you can get a 10 or 20%improvement from there, but
you're not going to get thiseight, you know, 2x every 18
months thing anymore. So eventhough I think that there's sort
of hope here for the next 10years, let's say very
interesting, if you can find itnow. Who's going to win? I would
say the companies that have thebest hardware and software
(47:47):
engineers that can understandwhat's needed in terms of
workloads and and the special,specialized processing hardware
architecture and softwarearchitecture. Those are the
companies that are going to winthat see that and do it. But
then after that 10 years, Idon't know, I don't know, I
don't know what's gonna happen,then I think maybe there'll be
some, you know, there could besome breakthrough in the
(48:08):
hardware side. There could be,you know, quantum
Chander Chawla (48:11):
community, yeah, that could
Mike Polacek (48:13):
be like self driving cars too
Chander Chawla (48:17):
far. This is so fascinating. You know, I've
known you for 10 plus years, butevery time we talk about these
things, I'm like, Oh, wow, thisis fascinating. So I think we
are at time we should end here.
Want to say any parting wordsMike, as more advice to
entrepreneurs or people wholisten to us, who are mainly in
(48:40):
the Nordic countries, and theyare mainly in the software
world. But I think there's lotof things that we can learn from
the semi world, thinking aboutmaking sense of ambiguity,
taking risk, finding markets. SoI think any pardoning words in
that.
Mike Polacek (49:01):
I mean, I would say the only thing I could think
of that's worthwhile to end on,maybe, is, is that there's, you
know, these companies don't havea monopoly on innovation. They
don't have a monopoly on greatproducts. All the companies that
you see today, you think, Oh, myGod. Like, how could anybody do
anything bigger, better, orwhatever, than them? And I've
(49:23):
seen it. I'm old enough now, andI've seen it so many times in my
career, where, you know, soon asa soon as you say that, then
somebody comes out withsomething better, yeah, and and
so, and it doesn't have to begiant companies. It doesn't have
to be, you know, take over theworld at the beginning, but, you
know, small teams of people cando really great things and and
(49:44):
really, you know, changemarkets, change change, change
the market environment on theirown. So, you know, be confident
that as you're doing the workthat you're doing, be bold. Be
you know, take chances and. And,and there's nothing saying that
that the next successful startuphas to be out of Palo Alto. It
(50:07):
can. It can be anywhere in theworld,
Chander Chawla (50:08):
yes. Well, thank you very much, Mike for joining
us, and thanks to our listenersfor joining us. Arne will be
back next week, and I will beback in Palo Alto. Thanks. Thank
you. Bye.
Popular Podcasts
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Therapy Gecko
An unlicensed lizard psychologist travels the universe talking to strangers about absolutely nothing. TO CALL THE GECKO: follow me on https://www.twitch.tv/lyleforever to get a notification for when I am taking calls. I am usually live Mondays, Wednesdays, and Fridays but lately a lot of other times too. I am a gecko.
The Joe Rogan Experience
The official podcast of comedian Joe Rogan.