August 28, 2025 • 32 mins
X-ray technology reveals what the naked eye cannot see - the intricate world of semiconductor interconnects that power our digital lives. In this deep dive with Ben Peecock, Senior Director of Business Development at Nordson Test and Inspection, we uncover the critical differences between x-ray inspection and metrology that keep semiconductor manufacturing on track.
With nearly three decades of experience in the industry, Peecock guides us through the evolution of semiconductor inspection technologies. While silicon itself remains transparent to x-rays, the metal interconnects between components provide the perfect canvas for quality assessment. As advanced packaging pushes toward smaller, more complex structures with 3D stacking and chiplet architectures, the need for sophisticated inspection has never been greater.
We explore how inspection (focused on imagery) and metrology (centered on precise measurements) serve complementary roles across the semiconductor manufacturing ecosystem. From R&D laboratories perfecting new processes to high-volume production lines seeking zero defects, these technologies help manufacturers identify issues before they become costly failures. The conversation ventures into the distinctions between 2D inspection (optimized for speed) and 3D analysis (delivering comprehensive structural information) and when each approach proves most valuable.
Particularly fascinating is Nordson's approach to vertical integration, developing their own specialized x-ray sources and detectors optimized specifically for semiconductor applications. This expertise extends to their innovative work integrating artificial intelligence to accelerate inspections while maintaining accuracy. Their thoughtful approach to data security gives customers options to protect proprietary information while still benefiting from AI's capabilities.
Discover how these technologies are already supporting emerging trends like panel-level packaging and learn about Nordson's unique radiation management solutions that protect sensitive components during inspection. Whether you're a semiconductor professional seeking quality control insights or simply curious about the technologies that ensure your electronic devices function reliably, this episode offers a fascinating glimpse into an invisible world of quality assurance.
Delivering best-in-class test, inspection, and metrology solutions for semiconductor applications.
Disclaimer: This post contains affiliate links. If you make a purchase, I may receive a commission at no extra cost to you.
Become a sustaining member!
Like what you hear? Follow us on LinkedIn and Twitter
Interested in reaching a qualified audience of microelectronics industry decision-makers? Invest in host-read advertisements, and promote your company in upcoming episodes. Contact Françoise von Trapp to learn more.
Interested in becoming a sponsor of the 3D InCites Podcast? Check out our 2024 Media Kit. Learn more about the 3D InCites Community and how you can become more involved.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Françoise von Trapp (00:00):
This episode of the 3D Insights
Podcast is sponsored by NordsenTest and Inspection.
Leaders in acoustic, opticaland x-ray inspection and
metrology systems for thesemiconductor and SMT
manufacturing markets.
Nordsen's next-generation 3Dx-ray inspection software,
dynamic Planar CT, revealsincredible details for Nordsen's
(00:20):
zero-defect strategy.
Nordsen's advanced software istwo times faster than planar CT,
with enhanced detail quality,superior reconstruction and a
larger field of view forincreased coverage and shorter
cycle times all to improveyields, processes and
productivity.
Learn more at Nordsoncom.
Hi there, I'm Francoise vonTrapp, and this is the 3D
(00:55):
Insights Podcast.
Hi everyone, welcome back tothe 3D Insights Podcast series
on x-ray inspection andmetrology for wafer level and
advanced packaging.
This is the third episode in aseries with our elite member
Nords Intestine Infection.
Now, in previous episode weexplored first the acoustic
(01:16):
microinspection and we talkedabout dynamic planar CT imaging.
Now in this episode we'retackling the differences between
x-ray inspection and metrology.
So here to help me is subjectmatter expert Ben Peecock.
Welcome to the podcast, ben.
Ben Peecock (01:32):
Hi, thank you.
Thanks for having us.
Nice to meet you.
Françoise von Trapp (01:35):
So I'm really excited for our
conversation today.
I think that this is one of themost interesting elements of
the advanced packaging space,because it's so critical.
But before we dive in, can youtell us about yourself and your
role at Nordson?
Ben Peecock (01:50):
Sure, I am currently the Senior Director of
Business Development for theNordson Test and Inspection
Group of Companies.
I've been with the business forover 29 years.
I was acquired by Nordson 20years ago, so yeah, a long
history.
I have been through thebusiness from the very earliest
days of X-Ray, kind of rightthrough.
(02:12):
I've worn a number of hats inmy time, starting in R&D as a
mechanical design engineer andthen going through operations,
finally through into businessdevelopment, where I am today.
Françoise von Trapp (02:23):
Okay, you know, in the earlier days of
advanced packaging and pleasecorrect me if I'm wrong it
seemed like x-ray inspection wasreally mostly for the R&D end
of things, not so much used inmanufacturing.
Has that changed?
Ben Peecock (02:38):
We definitely support all different areas of
the semiconductor industry, soright through from early R&D
development for new products,but also going through into
semiconductor volumemanufacturing and then right
through into final use.
So we also have equipmentthat's going right down to PCBA
(03:00):
or even some final assembled,tested products.
So testing at the full spectrumand again, that has been there
for some time.
Obviously, with every change intechnology there's a new
opportunity, and so we kind ofstart the cycle again, initially
with R&D and then followingthrough into volume production
(03:21):
as those products start to rampup.
Françoise von Trapp (03:24):
Okay, I think one of the reasons I
always latched on to x-rayinspection is it's something
that I could understand, havinghad x-rays on bones and what you
look for.
And I remember early onlearning that, okay, silicon
isn't something that you canx-ray, so that's not really what
you're looking at.
You're not looking for defectsin silicon when you're doing
these x-rays.
You're actually looking foreither anomalies or defects in
(03:48):
the actual interconnects.
Is that correct?
Ben Peecock (03:51):
Correct.
Yeah, as you say, silicon isvery translucent to x-rays, so
we can't see the silicon itself.
But, as you say, as we'removing into advanced packaging,
it's all about the interconnectsbetween the different devices,
the different layers, and that'swhere we have a really good
imaging from the x-ray to beable to see those, the bumps and
(04:11):
the tsvs and the variousinterconnects and and the
quality of those and thealignment of those and how
they're all stacked up andlooking for specific defects
within them.
We're using x-ray not just inthe way of medical x-ray where
we're looking through an object.
We're actually using x-ray as amicroscope, so using the
principle of geometricmagnification, by getting our
(04:33):
samples very close to the x-raysource, it then projects a large
view of the object underinspection.
So being able to see through it, being able to see a very high
magnification, it gives us theability to look inside these
advanced package devices.
Françoise von Trapp (04:49):
You mentioned TSVs and you mentioned
micro bumps.
Now, before that, 20 years ago,it was wire bond.
The bumps were bigger, thefeatures sizes were larger.
What are some of the trends anddynamics in wafer level and
advanced packaging today thatyou've seen shift over the years
?
Ben Peecock (05:09):
The reality now is that advanced packaging is
really driven by the high powercomputing, the whole AI world,
the need for very complexdevices starting to move from a
2D into a 3D space All thosereally driving the performance
of the device, meaning thatthose bumps and those
(05:29):
interconnects getting smallerand smaller, a lot more layers
being added.
So now the criticality aroundnot just is there a good
connection, but the alignment ofthose.
Are they all aligned?
Also, the fact that thesedevices now, because they are
getting more complex, they'regetting much higher value.
So finding defects early iscritical because obviously if
(05:51):
you've gone through the processand you're right down the end of
the chain and you've gotmultiple dyes all interconnected
and then you find a defect, thecost of failure is much higher.
So the importance of inspectionand that quality control is
even more critical.
Françoise von Trapp (06:07):
So are you seeing more use of x-ray
inspection than previously?
Ben Peecock (06:14):
Yeah, absolutely, as we have new technologies and
as the sort of global marketchanges and the need for
different countries'independence of their own supply
chain, that there's a lot ofparallel activities working on
developments of thesetechnologies.
So, again, not so much industrystandard lots of new ideas and
(06:36):
new processes being developed,and so lots of opportunities for
those processes needing to havequality control to make sure
they are solid when they ramp upinto high volume production.
Françoise von Trapp (06:50):
So one of the things we've been talking
about is x-ray inspection.
How is that different fromx-ray metrology?
Ben Peecock (06:57):
So typically x-ray inspection is looking at an
image, so the device isinspected under x-ray and an
image is produced, andoriginally it was all then an
operator reviewing that image tomake a determination of good
and bad product, and as time'sgone on that's become more
automated, but it's stillfundamentally based around an
(07:17):
image.
Metrology is actually aboutmeasurements, the, the physical
sizes of devices.
So using those same x-raytechniques, but actually using
it to give measurements, andthen the output of the tool
being data files rather than abank of images, and so it's not
requiring a human interventionto make a determination on good
(07:40):
or bad.
The data is the data which isthen typically fed back into the
customer's host computer tofeedback the quality of the
products and manage the qualitycontrol of their processes.
Françoise von Trapp (07:55):
What you're looking for is different and
from inspection you're lookingfor a bad, good and sorting
situation where you're removingand is metrology actually used
to refine processes?
Ben Peecock (08:09):
I mean they can both be used for refining
processes.
But metrology typically is avolume production mechanism.
It's all about how fast we canpush the process.
We can measure all of theproduct and identify any bad
parts within the product.
The inspection can be used inthe same way.
You have to determine whatyou're saying is a good and bad
(08:32):
product, whereas in metrologyyou're just getting measure the
size of the bump, measure thesize of the TSV, then measure
the size of the void, withinthat the fill level of the TSV,
size of the voids within that,the fill level of the tsv, and
then, of course, thedetermination can automatically
be applied that says anacceptable limit is a, is a void
of two microns or below, or apercentage fill of 98.
(08:54):
Is data driven as opposed toimage.
Françoise von Trapp (08:58):
Image output and so in the inspection
you said there is a manualoperator involved.
Ben Peecock (09:04):
That's where the industry started.
Everything was done with ahuman operator and as time has
moved on we've automated that.
So we're still taking images,but then automating some of
those measurement features.
The fundamental, the output, isan image, typically, then,
which will have an overlay,which will identify the good and
bad product and identify thevoids.
(09:24):
So then an operator can reviewthat data.
They can look at the imagethemselves and say the system
has determined this has got avoid of a certain size or it's
passed fail.
An operator can then look atthat image and make a
qualification.
So it doesn't need to have anoperator, but the images allow
that to be reviewed at a laterstage.
Françoise von Trapp (09:47):
So on the metrology side, you're measuring
the bumps and the voids and theTSVs.
Are you looking for anomaliesonly, or are you also?
I guess I'm trying tounderstand how that data is then
used.
Once you've gathered it, whatdo you do with the data?
Ben Peecock (10:03):
So it can be used for a whole range of different
inspections.
So from a simplest form, itmight be just the bumps on a
wafer.
So a 2D inspection how big arethe bumps?
Are there any missing bumps?
Are there any bumps that arejoined?
Are there any that are thewrong size?
And then what is the voiding?
Is there any air pockets insidethose solder bumps which could
(10:24):
then cause a problem later onthe process?
So that's kind of from thesimplest level.
But then we go into a full 3dwhere we might have a package
further down the line and we'renow looking at tsvs and a tsv
alignment with a micro bump.
Are those correctly positioned?
Are they off-center?
Is the shape and the sizegiving us a clue about the
(10:48):
quality of that interconnectjoint?
It's used both in the 2D andthe 3D space for, again, very
simple to highly complexspecific measurements which
might give us an indication of aspecific type of failure.
Françoise von Trapp (11:03):
Okay, so there's a lot of uses then for
x-ray metrology in the advancedpackaging and wafer-level
packaging space.
Ben Peecock (11:10):
Absolutely yeah.
Françoise von Trapp (11:12):
You know, basically from R&D all the way
through volume manufacturing.
Ben Peecock (11:16):
So typically we get involved with our customers
during product development.
So we will have some earlyprojects where they're looking
for specific applications andthen typically we then also put
those tools into the volumeproduction where they're looking
to manage their yields.
So in R&D it's about developinga product and a process.
(11:37):
When it then gets into fullproduction it's about managing
the yields and making sure theirproduction lines are operating
efficiently, they're not goingout of alignment or control that
need any any corrections, andmaking sure that the good
products identified and, wherethere is a where there is an
issue in the production line,making sure they're not spending
(11:59):
any more time adding value to aproduct which has already
failed.
So again, if you look at awafer which has got maybe
hundreds of dyes on there, itwill be able to identify those
which are bad and then feedingthat back to the process tools
to say, right, ignore thesespecific dyes because they have
a defect and there's no pointadding any more value, adding
(12:21):
any more process steps to those.
All the others are good andcontinue through to final
product.
Françoise von Trapp (12:28):
When, would you say, is the ideal time to
use 2D versus 3D?
Ben Peecock (12:32):
So typically 2D has the advantage of speed.
I mean, when we're talking 2D,what we're basically talking
about is a single image.
So we are either looking at astraight 2D straight through the
object, a bump from top down,and again, if that gives all the
information that's needed,that's by far the fastest way.
(12:54):
We can have a large areadetector.
We can inspect a high quantityof devices very fast.
When we start to get into the 3Dspace, this gives us the added
dimension which, in our, was tostart to look at those
interconnects in the verticalspace.
When you're then looking atTSVs, for example, or if you're
(13:15):
looking at interconnects, youneed to see not just straight
through but an oblique viewimage.
And if you're taking multipleoblique view images, we can then
turn that into a 3d model which, once you have then a model
constructed, then you canperform a sort of sectioning of
that model, removing parts ofthe device that are not clear or
(13:37):
get in the way so you canactually see the area of
interest.
So potentially having avertical slice through a TSV
which will show a lot moreinformation than just a plain
top-down view which you wouldget from a 2D inspection.
So typically 2D is all aboutspeed.
3d is when you need moreinformation about the inspection
point, which will take longerbut will give you a lot more
(13:59):
data.
Françoise von Trapp (14:00):
And that would be more in the development
side of things then.
Ben Peecock (14:04):
No, that can also be in full production, obviously
in full production.
The challenge is throughput.
Everybody's looking for thesetools to run as fast as possible
.
But yeah, there are someinspections which will always
need a 3D inspection point,whether it's in R&D or in full
volume production.
Françoise von Trapp (14:22):
I would imagine in full production that
you have to sample.
You can't inspect every singledevice that goes through.
Ben Peecock (14:29):
Again, depending on the customer and what they're
looking for, that's certainlyan option.
We do have some customers thatjust want to run batch
inspection.
They want to inspect fourpoints on a wafer just to make
sure.
In general, things areprogressing, but when they're a
high reliability, high valueproduct.
We're now seeing more drives todo 100 inspection.
But one of the things we'vebeen working on is is looking at
(14:50):
ways to get the benefits of 3dbut running at 2d line speed,
and there's been a real focus ofhow we integrate the artificial
intelligence into our equipmentto really leverage that,
because we have images andmodels of good and bad devices.
We can then use AI, train amodel to make a determination
(15:13):
and that's able to spot parts ofthe image in a 2D inspection,
which might give us a goodrepresentative of a 3D quality
inspection but at much fasterline speed, and so that's a big
area of focus for us right nowis how we're leveraging that AI.
Françoise von Trapp (15:31):
What are the different types of
inspection and metrologyplatforms available at Nordson
for x-ray metrology?
Ben Peecock (15:39):
When we started developing X-Ray products, we
really wanted to not just be anX-Ray company.
We wanted to be focused aroundthe electronics and
semiconductor space.
We recognized that thetechnology was underdeveloped
and so what we really focused onis that imaging chain the X-ray
(16:00):
source and the detector, whichare the key elements for
creating an x-ray image andreally bringing that design
in-house and then optimizing itaround our specific application
needs, the application needs foran electronic semiconductor
x-ray very being very differentfrom a medical x-ray.
So that was kind of where wecame from and that's always been
(16:21):
a key part of our technicaldifferentiation that we are
vertically integrated.
We own our own technology andwe're constantly driving to push
that forward to give us thevery highest quality imaging
chain in our particular space.
So what we've then done?
With that as the base platform,we've then built the different
systems, ranging right throughfrom a manual inspection tool,
(16:45):
where it really is all about anoperator using it, typically in
contract manufacturing, pcba,where they are looking for
something where they've got awide array of different customer
products and they want toinspect maybe the quality of the
bga device, soldering and anedge connector for solder fill,
(17:08):
um.
So starting at that entry levelon the manual inspection tools,
going right through into theautomated tools, which is all
about, again, typically PCBA orsome semiconductor device
inspection, and then rightthrough into the metrology tool,
(17:28):
which is the big waferinspection tool which, again,
we're now moving from not justinspecting wafers but also
moving to panel level inspectionas that industry starts to
develop.
Françoise von Trapp (17:42):
So covering the full spectrum, okay.
And when you say that you doeverything vertically, do you
develop then your own X-raysource?
Ben Peecock (17:50):
Yeah, so we've been manufacturing X-ray sources
around design for over 20 yearsand again we've optimized the
design specifically around thetype of devices we're inspecting
in our industry and again thatreally helps us get a very high
(18:12):
magnification and very highresolution X-ray source and then
on the detector side, a verylow noise, high resolution
detector, so we can get veryhigh magnification, very high
resolution images withrelatively low radiation dose.
Françoise von Trapp (18:25):
So the source is the same throughout
your systems, but the systemsare designed for the specific
applications, such as in the PCBand wafer and advanced
packaging space.
Can you talk a little bit aboutsome of specific application
examples, maybe share somespecifics from a customer
perspective?
Ben Peecock (18:45):
From sort of the entry level typically it's PCBA,
where we're doing void analysisin BGAs and solder fill volumes
in edge connectors In ourautomatic inspection lines,
typically again PCBA, but quiteoften they'll be for high
reliability applications, soaerospace, automotive,
(19:09):
defence-type applications wherereliability is critical.
So you might be doing the sametypes of inspection points but
you're doing 100% inspection andyou're doing it in very high
volumes.
And then through into the waferlevel inspection points, but
you're doing 100% inspection andyou're doing it in very high
volumes.
And then through into the waferlevel where we might be doing
(19:30):
the co-OS type products, whereagain looking at these advanced
packages, looking forinterconnect quality measurement
of voids in bumps and TSVvoiding and again a lot more
recently, on things like thealignment of the chip, so chip
gap height or chip alignmentwith the bumps.
Françoise von Trapp (19:51):
How about, like in the chiplet space?
Do you do anything there?
Or there's the embedded bridgesky.
Is that an application thatwould benefit from X-ray
metrology?
Ben Peecock (20:01):
We can cover any wafer-level product.
Again, we either do it at waferlevel or sometimes when it then
breaks out into singulateddevices.
Quite often we have somecustomers that are then
inspecting them at that lateststep where there's another layer
of interconnect being added andthey're looking for those final
inspection points and an inlinesystem.
Françoise von Trapp (20:24):
So you mentioned panel and panel level.
Packaging is a hot topic.
I don't know at this point howmuch is actually in production
at the panel level, but thenthere's also been a lot of talk
about the different substrates,like the glass panels versus
organic or other organic andsilicon.
So how do all of thesedevelopments impact your
(20:46):
decisions and your developmentprocess?
Ben Peecock (20:48):
to support that, so we work very closely with our
customers and, as we said,we're fortunate to be working in
both the production space butalso some of the R&D projects.
We've already developed andsupplied a panel version of our
metrology tool and we're workingon sort of expanding that range
in terms of different panelsizes and different materials as
(21:11):
they start to roll out.
As you say, it's a hot topic.
At the moment it's relativelysmall volumes.
So there's lots of activity inR&D, not so much in volume yet,
but we're certainly seeingindications that within the next
couple of years that will rampexponentially.
So at the moment, fairly smallbut significant growth coming.
(21:34):
And, as I say, we're able toreuse the same technology, the
same core technology, which wecontinue to develop.
We continue to generate thenext X-ray sources and detectors
to improve that resolution.
But fundamentally, the sametool then can be modified to
handle panels rather than wafers, just different robot loading,
(21:55):
different chucks and contactsfor those.
Françoise von Trapp (22:00):
Do you already have an automated panel
tool or are you developing thatas it evolves from R&D into
panel level?
Ben Peecock (22:10):
No, so we've already manufactured a version
of our metrology tool as a paneldevice and we have that.
It's up and running, it's at acustomer site, operating, and we
have multiple projectsexpanding on that with different
panel sizes and differentmaterials.
So, yeah, we already, wealready have a design out there
that's coming.
Françoise von Trapp (22:30):
So now is this your XM8000 AXM system that
we're talking about.
Ben Peecock (22:36):
Yeah, so that's when we talk about panel and
wafer level metrology.
It's the XM8000.
Françoise von Trapp (22:41):
Yeah, okay, and so is it the only solution,
or are there alternativesolutions on the market?
Ben Peecock (22:47):
There's not a lot of other products out there
doing the combination of thevery high resolution devices,
the clean room, compatibleautomatic handling for the
wafers and panels and thenaccelerating with AI.
So, in terms of all of thosekey elements, that there isn't
(23:08):
anything out there that iscompeting in that space.
Françoise von Trapp (23:11):
So can you talk about some of the
competitive advantages that itmight have in that space.
Ben Peecock (23:15):
So can you talk about some of the competitive
advantages that it might have.
Yes, because we control thatimaging chain, we are
manufacturing our own sources,again really optimizing it
around the samples we're lookingat.
So typically when you'relooking at higher resolution
x-ray sources, you're putting alot of energy into a very small
focal spot and the challengewith that is generating enough
(23:36):
X-rays to produce an image.
So because it's optimizedaround our products, we say
optimize that spot size and thenwe have the detector which can
then have the sensitivity topick up on a very low level of
X-ray photons to be able togenerate an image.
So controlling that imagingchain and really focusing on
(24:00):
this key market is what's kindof giving us the edge on the
image quality and, as I say,it's years of experience in how
we're processing those imagesand models to make measurements
and then make determinationabout good, bad, defects and
defect classification.
Françoise von Trapp (24:17):
And what role does software play in all
of this?
Because we've been talkingabout AI, and is AI a software
element?
Ben Peecock (24:24):
So the classical algorithms for determining from
an X-ray image the quality ofthe product is, again, it is
something we've built overdecades of experience and again,
we have a team of experts thatdeveloped a whole host of
algorithms for all sorts ofdifferent types of defects.
(24:45):
But what we're now doing isaccelerating that.
So, using that industryknowledge and that host of
algorithms we already have,using ai to accelerate that and
actually saying, okay, there aresome of those algorithms where,
where, if you have an imagewith multiple layers and
multiple objects in the field ofview, if we can then use ai to
actually determine which is theareas of interest, which which
(25:09):
layers of the bga are weactually wanting to inspect, so
a combination of the classicalgorithm with AI technology to
identify what we're trying toinspect, we can get either
faster throughput, more reliableon our results or even, as I
said before, using 2D inspectionwhere typically 3D would be the
(25:31):
more traditional way ofachieving that.
Ai by itself is not enough.
It needs to have thatbackground of enough.
It needs to have thatbackground of understanding.
Françoise von Trapp (25:37):
It needs to be trained right, right.
So AI is another hot topicright now, and we hear a lot
about generative AI, and we alsohear a lot about generative AI
not being accurate because ofthe data that it's receiving,
right?
So I'm assuming that when we'retalking about AI here, you're
actually working with closeddata sets, so the accuracy or
(26:01):
the intelligence is actuallymaybe smarter than what we're
getting just out on the internetwith generative AI.
Ben Peecock (26:08):
Yes, exactly Because we are controlling the
data that's being used to trainthe models, we can ensure that
the quality of that data is.
I mean, with the AI engines,they can just pull all that
information from the internetgood and bad, true and untrue
and so that's where we cansometimes lead to false results,
but where we have control ofthat, the training model, what
(26:32):
we're putting into it.
We also recognize that not allcustomers are happy for their
information to go into thosemodels or even to use that.
So we also offer the abilityfor customers to effectively
train their own models on site,having their own images that
they capture and classifythemselves, or they can update
(26:53):
and improve the quality of themodels that they're running.
Françoise von Trapp (26:57):
So that's a really good point, because if
you're training your tools onavailable customer data, there's
a lot of proprietaryinformation there that you can't
really use right.
You have to be really carefulabout you, know.
Yeah about.
Ben Peecock (27:18):
You know.
Yeah, yeah, absolutely, and Ithink that's.
There's a balance to be hadbetween the benefits of the
larger model having a moregeneral and higher capability
than a smaller niche specificmodel.
But again, depending on whatthe device is.
So a contract manufacturerusing their machine to just look
at the quality of inspection ofa BGA on a printed circuit
board, there's nothingproprietary about solder bumps
(27:40):
at that level.
Whereas if you're looking atadvanced packaging, where there
is a specific, unique processbeing developed by one
semiconductor manufacturer, thatmight be something that those
images won't leave the fab.
So we need to be able to coverthe full spectrum.
Françoise von Trapp (27:56):
Do you offer tools available with or
without AI?
Ben Peecock (28:00):
Yes, so our standard tools are available
with the classical algorithmsand again we're adding AI as an
additional bolt on and again,but on the basis that AI is
growing.
We have a number of featuresthat the AI is already
delivering for us today, but wehave a whole roadmap of
(28:20):
additional functionality thatwe're going to continue to add,
to build on.
So it's something that is notstatic, so we wanted to separate
that from the main software.
Françoise von Trapp (28:31):
I'll tell you where this question is
coming from.
I recently bought a new washerdryer and the last thing I need
on my washer dryer is AI.
But it's there because youcan't seem to get the other
things without it, you know.
So it's AI enabled and it'salso Wi-Fi.
Another thing I don't need onmy washer dryer is Wi-Fi, and
(28:53):
I'm assuming that this washerdryer is not going to last very
long because the electronicspanel is only warrantied for a
year.
So where I'm going with this isdo these tools actually need to
be connected to either anintranet or the internet to use
the AI?
Ben Peecock (29:12):
No.
So what we have is basicallythe runtime model, so the
software that's doing the AIpiece, that can be anywhere,
that can be on the cloud, thatcan be on a customer's central
server, it could even be on thesame PC of the machine.
The location of that is notcritical, obviously.
If you're connected to thecloud, then you have the ability
(29:33):
of the automatic updates andthat can happen a lot faster.
But it doesn't need to be, andcertainly that's that's.
One challenge that anybody inthe industry faces is that
people don't want to connectsome of their production tools
onto the cloud, and they arevery strictly prohibited.
So yeah, we're having theflexibility to be able to have
(29:53):
the runtime sitting directly onthe tool itself really gives us
that flexibility, okay.
Françoise von Trapp (29:59):
Is there anything else that we've missed
that you want to make surepeople understand?
Ben Peecock (30:04):
Yes, there is.
Another area of focus for us isradiation management.
So one of the challenges thathas been known about for a long
time is the potential damage ofradiation on certain
semiconductor devices, and so wehave spent a lot of time around
(30:25):
dose management and have anumber of different solutions,
whether it's about themonitoring and measurement of
that dosage, so we canaccurately report the level of
radiation that each componenthas received.
We can also use the same toolto effectively do that modelling
, so before we put a sample intoa machine we can model what the
(30:46):
dose expectation would be.
But we've also been focusing alot recently on how we can
provide automatic dynamicshielding for key areas of the
device.
So we know there are somecomponents which are more
sensitive to radiation thanothers.
So a big area of focus for usis a patented solution for
(31:08):
radiation dose shielding onwafer level devices, which
really opens up the capabilityof inspecting a lot more
products a lot further throughthe production chain.
Françoise von Trapp (31:20):
And is this unique to Norton?
Ben Peecock (31:23):
Yep.
So this is a Norton patentedtechnology and it's rolling out
now in live product todayExcellent.
Where can people go to learnmore?
Live product today, excellent.
Françoise von Trapp (31:33):
Where can people go to learn more?
Ben Peecock (31:35):
We're attending a number of trade shows.
We'll be in Semicon Taiwan,semicon West.
Obviously, our website's fullof information and we have our
full regional sales anddistribution service team that
can connect people to answer anyquestions they might have for a
range of demonstration.
Françoise von Trapp (31:54):
Excellent.
We will put some links in theshow notes and, yeah, this will
be out before SemiCon Taiwan, sothank you so much for joining
me today.
Ben Peecock (32:04):
Thank you very much.
Françoise von Trapp (32:10):
Next time on the 3D Insights podcast, I'm
joined by members of the IMAPScommittee to talk about the
upcoming IMAPS Symposium takingplace September 29th to October
30th.
There's lots more to come, sotune in next time to the 3D
Insights Podcast.
The 3D Insights Podcast is aproduction of 3D Insights LLC.
This episode of the 3D Insights
Podcast is sponsored by NordsenTest and Inspection.
Leaders in acoustic, opticaland x-ray inspection and
metrology systems for thesemiconductor and SMT
manufacturing markets.
Nordsen's next-generation 3Dx-ray inspection software,
dynamic Planar CT, revealsincredible details for Nordsen's
(00:20):
zero-defect strategy.
Nordsen's advanced software istwo times faster than planar CT,
with enhanced detail quality,superior reconstruction and a
larger field of view forincreased coverage and shorter
cycle times all to improveyields, processes and
productivity.
Learn more at Nordsoncom.
Hi there, I'm Francoise vonTrapp, and this is the 3D
(00:55):
Insights Podcast.
Hi everyone, welcome back tothe 3D Insights Podcast series
on x-ray inspection andmetrology for wafer level and
advanced packaging.
This is the third episode in aseries with our elite member
Nords Intestine Infection.
Now, in previous episode weexplored first the acoustic
(01:16):
microinspection and we talkedabout dynamic planar CT imaging.
Now in this episode we'retackling the differences between
x-ray inspection and metrology.
So here to help me is subjectmatter expert Ben Peecock.
Welcome to the podcast, ben.
Ben Peecock (01:32):
Hi, thank you.
Thanks for having us.
Nice to meet you.
Françoise von Trapp (01:35):
So I'm really excited for our
conversation today.
I think that this is one of themost interesting elements of
the advanced packaging space,because it's so critical.
But before we dive in, can youtell us about yourself and your
role at Nordson?
Ben Peecock (01:50):
Sure, I am currently the Senior Director of
Business Development for theNordson Test and Inspection
Group of Companies.
I've been with the business forover 29 years.
I was acquired by Nordson 20years ago, so yeah, a long
history.
I have been through thebusiness from the very earliest
days of X-Ray, kind of rightthrough.
(02:12):
I've worn a number of hats inmy time, starting in R&D as a
mechanical design engineer andthen going through operations,
finally through into businessdevelopment, where I am today.
Françoise von Trapp (02:23):
Okay, you know, in the earlier days of
advanced packaging and pleasecorrect me if I'm wrong it
seemed like x-ray inspection wasreally mostly for the R&D end
of things, not so much used inmanufacturing.
Has that changed?
Ben Peecock (02:38):
We definitely support all different areas of
the semiconductor industry, soright through from early R&D
development for new products,but also going through into
semiconductor volumemanufacturing and then right
through into final use.
So we also have equipmentthat's going right down to PCBA
(03:00):
or even some final assembled,tested products.
So testing at the full spectrumand again, that has been there
for some time.
Obviously, with every change intechnology there's a new
opportunity, and so we kind ofstart the cycle again, initially
with R&D and then followingthrough into volume production
(03:21):
as those products start to rampup.
Françoise von Trapp (03:24):
Okay, I think one of the reasons I
always latched on to x-rayinspection is it's something
that I could understand, havinghad x-rays on bones and what you
look for.
And I remember early onlearning that, okay, silicon
isn't something that you canx-ray, so that's not really what
you're looking at.
You're not looking for defectsin silicon when you're doing
these x-rays.
You're actually looking foreither anomalies or defects in
(03:48):
the actual interconnects.
Is that correct?
Ben Peecock (03:51):
Correct.
Yeah, as you say, silicon isvery translucent to x-rays, so
we can't see the silicon itself.
But, as you say, as we'removing into advanced packaging,
it's all about the interconnectsbetween the different devices,
the different layers, and that'swhere we have a really good
imaging from the x-ray to beable to see those, the bumps and
(04:11):
the tsvs and the variousinterconnects and and the
quality of those and thealignment of those and how
they're all stacked up andlooking for specific defects
within them.
We're using x-ray not just inthe way of medical x-ray where
we're looking through an object.
We're actually using x-ray as amicroscope, so using the
principle of geometricmagnification, by getting our
(04:33):
samples very close to the x-raysource, it then projects a large
view of the object underinspection.
So being able to see through it, being able to see a very high
magnification, it gives us theability to look inside these
advanced package devices.
Françoise von Trapp (04:49):
You mentioned TSVs and you mentioned
micro bumps.
Now, before that, 20 years ago,it was wire bond.
The bumps were bigger, thefeatures sizes were larger.
What are some of the trends anddynamics in wafer level and
advanced packaging today thatyou've seen shift over the years
?
Ben Peecock (05:09):
The reality now is that advanced packaging is
really driven by the high powercomputing, the whole AI world,
the need for very complexdevices starting to move from a
2D into a 3D space All thosereally driving the performance
of the device, meaning thatthose bumps and those
(05:29):
interconnects getting smallerand smaller, a lot more layers
being added.
So now the criticality aroundnot just is there a good
connection, but the alignment ofthose.
Are they all aligned?
Also, the fact that thesedevices now, because they are
getting more complex, they'regetting much higher value.
So finding defects early iscritical because obviously if
(05:51):
you've gone through the processand you're right down the end of
the chain and you've gotmultiple dyes all interconnected
and then you find a defect, thecost of failure is much higher.
So the importance of inspectionand that quality control is
even more critical.
Françoise von Trapp (06:07):
So are you seeing more use of x-ray
inspection than previously?
Ben Peecock (06:14):
Yeah, absolutely, as we have new technologies and
as the sort of global marketchanges and the need for
different countries'independence of their own supply
chain, that there's a lot ofparallel activities working on
developments of thesetechnologies.
So, again, not so much industrystandard lots of new ideas and
(06:36):
new processes being developed,and so lots of opportunities for
those processes needing to havequality control to make sure
they are solid when they ramp upinto high volume production.
Françoise von Trapp (06:50):
So one of the things we've been talking
about is x-ray inspection.
How is that different fromx-ray metrology?
Ben Peecock (06:57):
So typically x-ray inspection is looking at an
image, so the device isinspected under x-ray and an
image is produced, andoriginally it was all then an
operator reviewing that image tomake a determination of good
and bad product, and as time'sgone on that's become more
automated, but it's stillfundamentally based around an
(07:17):
image.
Metrology is actually aboutmeasurements, the, the physical
sizes of devices.
So using those same x-raytechniques, but actually using
it to give measurements, andthen the output of the tool
being data files rather than abank of images, and so it's not
requiring a human interventionto make a determination on good
(07:40):
or bad.
The data is the data which isthen typically fed back into the
customer's host computer tofeedback the quality of the
products and manage the qualitycontrol of their processes.
Françoise von Trapp (07:55):
What you're looking for is different and
from inspection you're lookingfor a bad, good and sorting
situation where you're removingand is metrology actually used
to refine processes?
Ben Peecock (08:09):
I mean they can both be used for refining
processes.
But metrology typically is avolume production mechanism.
It's all about how fast we canpush the process.
We can measure all of theproduct and identify any bad
parts within the product.
The inspection can be used inthe same way.
You have to determine whatyou're saying is a good and bad
(08:32):
product, whereas in metrologyyou're just getting measure the
size of the bump, measure thesize of the TSV, then measure
the size of the void, withinthat the fill level of the TSV,
size of the voids within that,the fill level of the tsv, and
then, of course, thedetermination can automatically
be applied that says anacceptable limit is a, is a void
of two microns or below, or apercentage fill of 98.
(08:54):
Is data driven as opposed toimage.
Françoise von Trapp (08:58):
Image output and so in the inspection
you said there is a manualoperator involved.
Ben Peecock (09:04):
That's where the industry started.
Everything was done with ahuman operator and as time has
moved on we've automated that.
So we're still taking images,but then automating some of
those measurement features.
The fundamental, the output, isan image, typically, then,
which will have an overlay,which will identify the good and
bad product and identify thevoids.
(09:24):
So then an operator can reviewthat data.
They can look at the imagethemselves and say the system
has determined this has got avoid of a certain size or it's
passed fail.
An operator can then look atthat image and make a
qualification.
So it doesn't need to have anoperator, but the images allow
that to be reviewed at a laterstage.
Françoise von Trapp (09:47):
So on the metrology side, you're measuring
the bumps and the voids and theTSVs.
Are you looking for anomaliesonly, or are you also?
I guess I'm trying tounderstand how that data is then
used.
Once you've gathered it, whatdo you do with the data?
Ben Peecock (10:03):
So it can be used for a whole range of different
inspections.
So from a simplest form, itmight be just the bumps on a
wafer.
So a 2D inspection how big arethe bumps?
Are there any missing bumps?
Are there any bumps that arejoined?
Are there any that are thewrong size?
And then what is the voiding?
Is there any air pockets insidethose solder bumps which could
(10:24):
then cause a problem later onthe process?
So that's kind of from thesimplest level.
But then we go into a full 3dwhere we might have a package
further down the line and we'renow looking at tsvs and a tsv
alignment with a micro bump.
Are those correctly positioned?
Are they off-center?
Is the shape and the sizegiving us a clue about the
(10:48):
quality of that interconnectjoint?
It's used both in the 2D andthe 3D space for, again, very
simple to highly complexspecific measurements which
might give us an indication of aspecific type of failure.
Françoise von Trapp (11:03):
Okay, so there's a lot of uses then for
x-ray metrology in the advancedpackaging and wafer-level
packaging space.
Ben Peecock (11:10):
Absolutely yeah.
Françoise von Trapp (11:12):
You know, basically from R&D all the way
through volume manufacturing.
Ben Peecock (11:16):
So typically we get involved with our customers
during product development.
So we will have some earlyprojects where they're looking
for specific applications andthen typically we then also put
those tools into the volumeproduction where they're looking
to manage their yields.
So in R&D it's about developinga product and a process.
(11:37):
When it then gets into fullproduction it's about managing
the yields and making sure theirproduction lines are operating
efficiently, they're not goingout of alignment or control that
need any any corrections, andmaking sure that the good
products identified and, wherethere is a where there is an
issue in the production line,making sure they're not spending
(11:59):
any more time adding value to aproduct which has already
failed.
So again, if you look at awafer which has got maybe
hundreds of dyes on there, itwill be able to identify those
which are bad and then feedingthat back to the process tools
to say, right, ignore thesespecific dyes because they have
a defect and there's no pointadding any more value, adding
(12:21):
any more process steps to those.
All the others are good andcontinue through to final
product.
Françoise von Trapp (12:28):
When, would you say, is the ideal time to
use 2D versus 3D?
Ben Peecock (12:32):
So typically 2D has the advantage of speed.
I mean, when we're talking 2D,what we're basically talking
about is a single image.
So we are either looking at astraight 2D straight through the
object, a bump from top down,and again, if that gives all the
information that's needed,that's by far the fastest way.
(12:54):
We can have a large areadetector.
We can inspect a high quantityof devices very fast.
When we start to get into the 3Dspace, this gives us the added
dimension which, in our, was tostart to look at those
interconnects in the verticalspace.
When you're then looking atTSVs, for example, or if you're
(13:15):
looking at interconnects, youneed to see not just straight
through but an oblique viewimage.
And if you're taking multipleoblique view images, we can then
turn that into a 3d model which, once you have then a model
constructed, then you canperform a sort of sectioning of
that model, removing parts ofthe device that are not clear or
(13:37):
get in the way so you canactually see the area of
interest.
So potentially having avertical slice through a TSV
which will show a lot moreinformation than just a plain
top-down view which you wouldget from a 2D inspection.
So typically 2D is all aboutspeed.
3d is when you need moreinformation about the inspection
point, which will take longerbut will give you a lot more
(13:59):
data.
Françoise von Trapp (14:00):
And that would be more in the development
side of things then.
Ben Peecock (14:04):
No, that can also be in full production, obviously
in full production.
The challenge is throughput.
Everybody's looking for thesetools to run as fast as possible
.
But yeah, there are someinspections which will always
need a 3D inspection point,whether it's in R&D or in full
volume production.
Françoise von Trapp (14:22):
I would imagine in full production that
you have to sample.
You can't inspect every singledevice that goes through.
Ben Peecock (14:29):
Again, depending on the customer and what they're
looking for, that's certainlyan option.
We do have some customers thatjust want to run batch
inspection.
They want to inspect fourpoints on a wafer just to make
sure.
In general, things areprogressing, but when they're a
high reliability, high valueproduct.
We're now seeing more drives todo 100 inspection.
But one of the things we'vebeen working on is is looking at
(14:50):
ways to get the benefits of 3dbut running at 2d line speed,
and there's been a real focus ofhow we integrate the artificial
intelligence into our equipmentto really leverage that,
because we have images andmodels of good and bad devices.
We can then use AI, train amodel to make a determination
(15:13):
and that's able to spot parts ofthe image in a 2D inspection,
which might give us a goodrepresentative of a 3D quality
inspection but at much fasterline speed, and so that's a big
area of focus for us right nowis how we're leveraging that AI.
Françoise von Trapp (15:31):
What are the different types of
inspection and metrologyplatforms available at Nordson
for x-ray metrology?
Ben Peecock (15:39):
When we started developing X-Ray products, we
really wanted to not just be anX-Ray company.
We wanted to be focused aroundthe electronics and
semiconductor space.
We recognized that thetechnology was underdeveloped
and so what we really focused onis that imaging chain the X-ray
(16:00):
source and the detector, whichare the key elements for
creating an x-ray image andreally bringing that design
in-house and then optimizing itaround our specific application
needs, the application needs foran electronic semiconductor
x-ray very being very differentfrom a medical x-ray.
So that was kind of where wecame from and that's always been
(16:21):
a key part of our technicaldifferentiation that we are
vertically integrated.
We own our own technology andwe're constantly driving to push
that forward to give us thevery highest quality imaging
chain in our particular space.
So what we've then done?
With that as the base platform,we've then built the different
systems, ranging right throughfrom a manual inspection tool,
(16:45):
where it really is all about anoperator using it, typically in
contract manufacturing, pcba,where they are looking for
something where they've got awide array of different customer
products and they want toinspect maybe the quality of the
bga device, soldering and anedge connector for solder fill,
(17:08):
um.
So starting at that entry levelon the manual inspection tools,
going right through into theautomated tools, which is all
about, again, typically PCBA orsome semiconductor device
inspection, and then rightthrough into the metrology tool,
(17:28):
which is the big waferinspection tool which, again,
we're now moving from not justinspecting wafers but also
moving to panel level inspectionas that industry starts to
develop.
Françoise von Trapp (17:42):
So covering the full spectrum, okay.
And when you say that you doeverything vertically, do you
develop then your own X-raysource?
Ben Peecock (17:50):
Yeah, so we've been manufacturing X-ray sources
around design for over 20 yearsand again we've optimized the
design specifically around thetype of devices we're inspecting
in our industry and again thatreally helps us get a very high
(18:12):
magnification and very highresolution X-ray source and then
on the detector side, a verylow noise, high resolution
detector, so we can get veryhigh magnification, very high
resolution images withrelatively low radiation dose.
Françoise von Trapp (18:25):
So the source is the same throughout
your systems, but the systemsare designed for the specific
applications, such as in the PCBand wafer and advanced
packaging space.
Can you talk a little bit aboutsome of specific application
examples, maybe share somespecifics from a customer
perspective?
Ben Peecock (18:45):
From sort of the entry level typically it's PCBA,
where we're doing void analysisin BGAs and solder fill volumes
in edge connectors In ourautomatic inspection lines,
typically again PCBA, but quiteoften they'll be for high
reliability applications, soaerospace, automotive,
(19:09):
defence-type applications wherereliability is critical.
So you might be doing the sametypes of inspection points but
you're doing 100% inspection andyou're doing it in very high
volumes.
And then through into the waferlevel inspection points, but
you're doing 100% inspection andyou're doing it in very high
volumes.
And then through into the waferlevel where we might be doing
(19:30):
the co-OS type products, whereagain looking at these advanced
packages, looking forinterconnect quality measurement
of voids in bumps and TSVvoiding and again a lot more
recently, on things like thealignment of the chip, so chip
gap height or chip alignmentwith the bumps.
Françoise von Trapp (19:51):
How about, like in the chiplet space?
Do you do anything there?
Or there's the embedded bridgesky.
Is that an application thatwould benefit from X-ray
metrology?
Ben Peecock (20:01):
We can cover any wafer-level product.
Again, we either do it at waferlevel or sometimes when it then
breaks out into singulateddevices.
Quite often we have somecustomers that are then
inspecting them at that lateststep where there's another layer
of interconnect being added andthey're looking for those final
inspection points and an inlinesystem.
Françoise von Trapp (20:24):
So you mentioned panel and panel level.
Packaging is a hot topic.
I don't know at this point howmuch is actually in production
at the panel level, but thenthere's also been a lot of talk
about the different substrates,like the glass panels versus
organic or other organic andsilicon.
So how do all of thesedevelopments impact your
(20:46):
decisions and your developmentprocess?
Ben Peecock (20:48):
to support that, so we work very closely with our
customers and, as we said,we're fortunate to be working in
both the production space butalso some of the R&D projects.
We've already developed andsupplied a panel version of our
metrology tool and we're workingon sort of expanding that range
in terms of different panelsizes and different materials as
(21:11):
they start to roll out.
As you say, it's a hot topic.
At the moment it's relativelysmall volumes.
So there's lots of activity inR&D, not so much in volume yet,
but we're certainly seeingindications that within the next
couple of years that will rampexponentially.
So at the moment, fairly smallbut significant growth coming.
(21:34):
And, as I say, we're able toreuse the same technology, the
same core technology, which wecontinue to develop.
We continue to generate thenext X-ray sources and detectors
to improve that resolution.
But fundamentally, the sametool then can be modified to
handle panels rather than wafers, just different robot loading,
(21:55):
different chucks and contactsfor those.
Françoise von Trapp (22:00):
Do you already have an automated panel
tool or are you developing thatas it evolves from R&D into
panel level?
Ben Peecock (22:10):
No, so we've already manufactured a version
of our metrology tool as a paneldevice and we have that.
It's up and running, it's at acustomer site, operating, and we
have multiple projectsexpanding on that with different
panel sizes and differentmaterials.
So, yeah, we already, wealready have a design out there
that's coming.
Françoise von Trapp (22:30):
So now is this your XM8000 AXM system that
we're talking about.
Ben Peecock (22:36):
Yeah, so that's when we talk about panel and
wafer level metrology.
It's the XM8000.
Françoise von Trapp (22:41):
Yeah, okay, and so is it the only solution,
or are there alternativesolutions on the market?
Ben Peecock (22:47):
There's not a lot of other products out there
doing the combination of thevery high resolution devices,
the clean room, compatibleautomatic handling for the
wafers and panels and thenaccelerating with AI.
So, in terms of all of thosekey elements, that there isn't
(23:08):
anything out there that iscompeting in that space.
Françoise von Trapp (23:11):
So can you talk about some of the
competitive advantages that itmight have in that space.
Ben Peecock (23:15):
So can you talk about some of the competitive
advantages that it might have.
Yes, because we control thatimaging chain, we are
manufacturing our own sources,again really optimizing it
around the samples we're lookingat.
So typically when you'relooking at higher resolution
x-ray sources, you're putting alot of energy into a very small
focal spot and the challengewith that is generating enough
(23:36):
X-rays to produce an image.
So because it's optimizedaround our products, we say
optimize that spot size and thenwe have the detector which can
then have the sensitivity topick up on a very low level of
X-ray photons to be able togenerate an image.
So controlling that imagingchain and really focusing on
(24:00):
this key market is what's kindof giving us the edge on the
image quality and, as I say,it's years of experience in how
we're processing those imagesand models to make measurements
and then make determinationabout good, bad, defects and
defect classification.
Françoise von Trapp (24:17):
And what role does software play in all
of this?
Because we've been talkingabout AI, and is AI a software
element?
Ben Peecock (24:24):
So the classical algorithms for determining from
an X-ray image the quality ofthe product is, again, it is
something we've built overdecades of experience and again,
we have a team of experts thatdeveloped a whole host of
algorithms for all sorts ofdifferent types of defects.
(24:45):
But what we're now doing isaccelerating that.
So, using that industryknowledge and that host of
algorithms we already have,using ai to accelerate that and
actually saying, okay, there aresome of those algorithms where,
where, if you have an imagewith multiple layers and
multiple objects in the field ofview, if we can then use ai to
actually determine which is theareas of interest, which which
(25:09):
layers of the bga are weactually wanting to inspect, so
a combination of the classicalgorithm with AI technology to
identify what we're trying toinspect, we can get either
faster throughput, more reliableon our results or even, as I
said before, using 2D inspectionwhere typically 3D would be the
(25:31):
more traditional way ofachieving that.
Ai by itself is not enough.
It needs to have thatbackground of enough.
It needs to have thatbackground of understanding.
Françoise von Trapp (25:37):
It needs to be trained right, right.
So AI is another hot topicright now, and we hear a lot
about generative AI, and we alsohear a lot about generative AI
not being accurate because ofthe data that it's receiving,
right?
So I'm assuming that when we'retalking about AI here, you're
actually working with closeddata sets, so the accuracy or
(26:01):
the intelligence is actuallymaybe smarter than what we're
getting just out on the internetwith generative AI.
Ben Peecock (26:08):
Yes, exactly Because we are controlling the
data that's being used to trainthe models, we can ensure that
the quality of that data is.
I mean, with the AI engines,they can just pull all that
information from the internetgood and bad, true and untrue
and so that's where we cansometimes lead to false results,
but where we have control ofthat, the training model, what
(26:32):
we're putting into it.
We also recognize that not allcustomers are happy for their
information to go into thosemodels or even to use that.
So we also offer the abilityfor customers to effectively
train their own models on site,having their own images that
they capture and classifythemselves, or they can update
(26:53):
and improve the quality of themodels that they're running.
Françoise von Trapp (26:57):
So that's a really good point, because if
you're training your tools onavailable customer data, there's
a lot of proprietaryinformation there that you can't
really use right.
You have to be really carefulabout you, know.
Yeah about.
Ben Peecock (27:18):
You know.
Yeah, yeah, absolutely, and Ithink that's.
There's a balance to be hadbetween the benefits of the
larger model having a moregeneral and higher capability
than a smaller niche specificmodel.
But again, depending on whatthe device is.
So a contract manufacturerusing their machine to just look
at the quality of inspection ofa BGA on a printed circuit
board, there's nothingproprietary about solder bumps
(27:40):
at that level.
Whereas if you're looking atadvanced packaging, where there
is a specific, unique processbeing developed by one
semiconductor manufacturer, thatmight be something that those
images won't leave the fab.
So we need to be able to coverthe full spectrum.
Françoise von Trapp (27:56):
Do you offer tools available with or
without AI?
Ben Peecock (28:00):
Yes, so our standard tools are available
with the classical algorithmsand again we're adding AI as an
additional bolt on and again,but on the basis that AI is
growing.
We have a number of featuresthat the AI is already
delivering for us today, but wehave a whole roadmap of
(28:20):
additional functionality thatwe're going to continue to add,
to build on.
So it's something that is notstatic, so we wanted to separate
that from the main software.
Françoise von Trapp (28:31):
I'll tell you where this question is
coming from.
I recently bought a new washerdryer and the last thing I need
on my washer dryer is AI.
But it's there because youcan't seem to get the other
things without it, you know.
So it's AI enabled and it'salso Wi-Fi.
Another thing I don't need onmy washer dryer is Wi-Fi, and
(28:53):
I'm assuming that this washerdryer is not going to last very
long because the electronicspanel is only warrantied for a
year.
So where I'm going with this isdo these tools actually need to
be connected to either anintranet or the internet to use
the AI?
Ben Peecock (29:12):
No.
So what we have is basicallythe runtime model, so the
software that's doing the AIpiece, that can be anywhere,
that can be on the cloud, thatcan be on a customer's central
server, it could even be on thesame PC of the machine.
The location of that is notcritical, obviously.
If you're connected to thecloud, then you have the ability
(29:33):
of the automatic updates andthat can happen a lot faster.
But it doesn't need to be, andcertainly that's that's.
One challenge that anybody inthe industry faces is that
people don't want to connectsome of their production tools
onto the cloud, and they arevery strictly prohibited.
So yeah, we're having theflexibility to be able to have
(29:53):
the runtime sitting directly onthe tool itself really gives us
that flexibility, okay.
Françoise von Trapp (29:59):
Is there anything else that we've missed
that you want to make surepeople understand?
Ben Peecock (30:04):
Yes, there is.
Another area of focus for us isradiation management.
So one of the challenges thathas been known about for a long
time is the potential damage ofradiation on certain
semiconductor devices, and so wehave spent a lot of time around
(30:25):
dose management and have anumber of different solutions,
whether it's about themonitoring and measurement of
that dosage, so we canaccurately report the level of
radiation that each componenthas received.
We can also use the same toolto effectively do that modelling
, so before we put a sample intoa machine we can model what the
(30:46):
dose expectation would be.
But we've also been focusing alot recently on how we can
provide automatic dynamicshielding for key areas of the
device.
So we know there are somecomponents which are more
sensitive to radiation thanothers.
So a big area of focus for usis a patented solution for
(31:08):
radiation dose shielding onwafer level devices, which
really opens up the capabilityof inspecting a lot more
products a lot further throughthe production chain.
Françoise von Trapp (31:20):
And is this unique to Norton?
Ben Peecock (31:23):
Yep.
So this is a Norton patentedtechnology and it's rolling out
now in live product todayExcellent.
Where can people go to learnmore?
Live product today, excellent.
Françoise von Trapp (31:33):
Where can people go to learn more?
Ben Peecock (31:35):
We're attending a number of trade shows.
We'll be in Semicon Taiwan,semicon West.
Obviously, our website's fullof information and we have our
full regional sales anddistribution service team that
can connect people to answer anyquestions they might have for a
range of demonstration.
Françoise von Trapp (31:54):
Excellent.
We will put some links in theshow notes and, yeah, this will
be out before SemiCon Taiwan, sothank you so much for joining
me today.
Ben Peecock (32:04):
Thank you very much.
Françoise von Trapp (32:10):
Next time on the 3D Insights podcast, I'm
joined by members of the IMAPScommittee to talk about the
upcoming IMAPS Symposium takingplace September 29th to October
30th.
There's lots more to come, sotune in next time to the 3D
Insights Podcast.
The 3D Insights Podcast is aproduction of 3D Insights LLC.
Popular Podcasts
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
NFL Daily with Gregg Rosenthal
Gregg Rosenthal and a rotating crew of elite NFL Media co-hosts, including Patrick Claybon, Colleen Wolfe, Steve Wyche, Nick Shook and Jourdan Rodrigue of The Athletic get you caught up daily on all the NFL news and analysis you need to be smarter and funnier than your friends.