It's something that we call artificial
intelligence process control.

With the goal that I
mentioned the entire time

is scaling all the inventions that you
may have here and at other places.

Our ideas behind how this might be done
and how we're starting to do it and

how you know customers and
partners of ours are starting to do it.

Now this is this is
something that you may or

may not agree with this part we
think that life kind of goes

intentions of making the world a better
place I think nearly universal.

But even with those
intentions technologically.

Or even just society in general
you can be running parallel and

everything can be going fine but

they can start to diverging if we
start to choose the wrong things.

And I'll talk about what some of those
are I think are labels got lost here.

So I won't go into that too much I don't
know how that happened but we will the.

And we'll see what we think is
the right thing in the wrong thing.

And of course this is up for debate and
would love to have debate on it so

this is one called the Gigafactory that.

Tesla opened recently in Nevada.

It's an interest remotely
a large factory it's a single

floor extremely large factory for
making extremely small batteries.

Now if we go back in time and
I think part of that.

Part of that divergence was had to do
with you know making a very large batch.

Things in big facilities which we
consider kind of an old idea or

making in this very small and

affordable ie the way that somebody
like Eric Drexler described with with

using you know assemblers or even like
we're seeing with industrial three D.

printers where you're making something
that is about the size of the three D.

printer so
why have such an enormous facility.

I think that Tesla and
all of the rescue rest of

mosques companies have done amazing
things this is one idea I don't

think is really the way to look
at things in the future so

they do combined these ideas that
are starting to be considered the State

of the art you know normal things so
STATE OF THE AIR You know you have.

Really the state of the art which is the
we call the state of the art that bottom

part of the curve you know
what the state of the art

which is the state of the art
is also status and in our view.

The what we want is improvement so
this is the way factories have been done.

There's a there is a natural leg from lab
to factory and something like this occurs.

The Internet of things we talked a bit
I.O.T. doc just was talking about

it's amazing and important we could
think of ourselves in a way of being

an industrial Internet of Things company
but in some ways it's a dangerous concept.

In the sense that you're just
producing a lot of data and

what do you do with that data if you
have data across this whole facility

you have to figure out how to combine that
data and get rid of the wasted data so

the data being a type
of pollution in a sense

unless you know how to manage that
data which we'll get into a bit.

I'm.

Distributed networks instead of
vertically integrated networks

something else we'll talk about.

So in intended future we're
going to see which is

our are our ideas that of
course are not only our ideas.

I've learned that don't be
careful about taking credit for

anything because somebody in this room
might be doing the same thing and

I welcome that it's wonderful if people
are working on similar things where you

have process optimization high yield less
waste this is the this is the future that

is that divergence is going up you might
think of really these are two conventions

in a way the Hyperloop and they get
a factory but one we consider to be

a vertically integrated idea that affects
transportation things like this so.

So you saw one big picture
of our instrument but

part of this is that it isn't taking
up a large part of an enormous factory.

It's also can be so it can also be
right next to a production tool and

I'm repeating some of what
I mentioned earlier but

we'll just go into a bit
more detail about that.

So it's moving in this case
a semiconductor wafer moving it

onto a macro inspection aligning it and

then doing inspection on
a smaller scale it's one tool.

This is something that John will
talk about an application later on

in some of being egged Gnostic to
applications this is an application for

a film that goes underneath the screen
of an i Phone Is that correct so

this type of technology that
is critical seems like.

An old material science
invention in a way but

if you're dealing with
something that requires it.

Pixel is a pixel that's wrong
can propagate through out.

A screen this is an issue so using a tool

online with a roll the roll manufacturing
and still be able to do inspection.

Work something I give this
rather than that small or

this is still incredibly small for
materials lab.

But it's you know what we roll
to roll doing the same thing

this is actually a conceptual design
we have done for an aerospace company

that does carbon composites for
parts of an airplane where they have to

overlay these manually overlay
the different layers of the carbon

composite at angles that
are very are very specific and

some of the things that we're looking
at could not be more mission critical

we we work with you know if you're
working on a self driving car for

instance you can't get anything wrong so
you need good AI For this you need good

you need to make sure that everything
comes out perfect obviously an airplane

you want to make sure that a human that's
inspecting it does not get anything wrong

so this actually does not eliminate humans
they move to different days lay them down

imaging they they inspect
train an AI system and

then place it down and
move it to others so it makes those jobs

not the Boreas miserable jobs that we had
said before but makes it something that

produces quantitative interesting
results that are guided by a neural net.

But still involves people so

it looks something like this
this is conceptual work but

it can work even over you know using this
boom to look at an entire you know for

oil in an airplane something
called telescopic microscopy so

my cost me a bit looking
at over large areas.

Yes the whole wing.

So Tories and intended future that.

A lot of this is the world that we're
we're starting in with partners to

you know do little bits and pieces of but
where does this all go where does it go

from imaging right next to
something to creating cities so

I think this is really grand vision
things some of them will see

sooner rather than later I think I I'm
hoping that we see them all eventually.

So what this is this is how you
can think of a normal factory.

And it doesn't matter if it's the most
sophisticated semiconductor factory or

a car factory I say that if Henry Ford
were to come back right now and

walk through a fab.

Course he would have been injured a clean
room couldn't imagine the electronics that

are involved but
the process is still his assembly line.

Things still work in that map manner and
there's some inherent problems with that

as you move to more critical
things going forward.

A big part of that is that it is simply
a line is meant to produce the same

thing over and over again that's good
right you know if you set a spec

you continue to produce it over and
over again but

why though if you're going to keep
up with doing rapid iterations and

improving quality why that has its
limitations so this just goes on and

on until you have something that.

Looks like this and
you're getting to be a bigger and

bigger factor to something that looks
like this and that's agenda up with.

She is I can't believe I mean
that's critical of this factory

This is a pretty I'm a.

Easing invention in there's solar panels
across the entire thing the power it it's

hopefully will produce the best batteries
possible and it very well might.

But just to push back on not
it's easy to push back on

what we consider is a low margin old
fashioned business is it's less that

people are pushing back on things like
this but there are reasons to I think.

Now this is history that you may all know
but something that we keep in mind a lot.

Before World War two I would say so

really in the night we talk about this
leg between inventions in labs versus

things that happened in production and so
in the late twenty's out of Bell Labs.

And inventors came up with an idea
of statistical process control

where because you didn't have sensors that
can measure and produce enormous amounts

of data there was this is to Stickley
relevant amount of data that would allow

you to have a process that's in control
whether then what happened before that

which was making sure that at either steps
in the process or at the end of a process

of production line you would have just
pass or fail quality control checks.

So that that invention incurred but then
there became this necessity post World War

two where Deming went to Japan and
basically rebuilt

a country that had been burned to
the ground and ended up making you know

taking this some of these early Bell Lab
ideas putting them into practice so

that you get these amazing Chinese
production ideals that we are still using

in the world of S.P.C. so things like lean
manufacturing and it's all based on S..

See.

One way you can Fink of this is that

you could imagine doing inspection
physical inspection on you know it's

a designed experiment that has some
type of statistical relevance and

that would produce a product is
that what meets your spec you so

people that well that that's old because
now we're collecting a lot of data but

even though we're collecting a lot of data

that data is limited to what
we're collecting it from.

So your the idea of what dimming
did be to create control charts for

each part of the process so something is
significantly different in the mean it's

a certain amount of standard deviations
away from the mean it's not it's not

falling out of quality you know
it would still pass quality but

the process would be
altering being altered so

that you know that there's an issue
in that part of the process.

Basic idea it's sampling it's interesting
the same thing occurs even if you have

sensors at each of those levels so I put a
little bit back on Internet of Things and

this is done correctly so you could
have a part of a process than if you're

familiar something called P ID
controllers this is something existed for

a long time that on say a temperature
controller you set you have a set point

of a temperature that you would like to
get to and there's some simple regression

analysis that happens to make sure that
it always stays within that control so

in a way that sort of this a small
AI A P I D controller that allows for

S.P.C. to be allows
Deming's model to work but

if you have a process that is bit as
large as what we're talking about with

a semiconductor fab or
that Gigafactory or other things

what happens at the end of the entire
process even if you're not out of those.

Limits on any individual

layer of the process you can
have a jittery final product.

So there's no speaking back and
forth between the process itself and

different parts of the process and
the whole not in

a I doubt it's not how some go by
propagation works where there is

you know you're assigning causality along
the way because your computer you're

communicating back to a central system
that's saying wait it doesn't matter what

the temperature was because somehow three
steps later something else went wrong so

maybe the temperature should be altered
even if it was out of what we thought

the temperature should be so in the end
you have this self-contained system.

I mean do you mind body is like this
things go wrong all the time but

it's self correcting.

And if you have good analytical tools
you can start to assign causality to

why something happened why an infectious
disease was caused by a mosquito bite and

so on this is a similar thing
that we think factories should be

they should not be just the dimming
model but should be it's not S.P.C. but

a I.P.C. so dimming did a great job but
we want to actually every time get better.

Our factories should be better
our products should be better and

that information of where causality were
something could have gone out of control

you can and you can address it and
you can invent something new.

So.

If we think about.

This non vertically integrated
world we live in where

in the semiconductor world there
are a lot of companies that in

I think starting in the eighty's and
ninety's went to some fabulous fabs where.

He would go to a foundry by and
this still happens all the time you buy.

Your way for
from one place to integrated another and

you have a sort of a large supply
chain this is great unifies the world.

In a way this is great commerce

it doesn't necessarily allow for
something like a I.P.C..

So I mean if we look by the way in a we
inspect many layers of an i Phone for

instance and we're talking about
factories in Cupertino chipsets

that are made in Italy and another
company assembled at Foxconn in China.

So if there's something wrong with the
coding versus something wrong with this

that this screen versus something that's
wrong with the chipset the really you're

dealing with some statistical variations
within each node being a whole factory

not a node even being parts of the process
so this is sort of spiral thing.

On the far side is what we
consider something just

moving spiraling down a single
factory that can make something.

What's what's interesting is this also
comes out of a Japanese ideology so

I said maybe certain things don't
apply anymore because we have so

much data and we have and
you know we have deep learning and

basing probabilistic learning and
all of these other tools available to us

one thing that did work in something
called lean manufacturing.

Was this idea that you
don't necessarily want to

have the largest batch now semiconductor
companies don't do this they

make they go from you
know a one inch wafer

that has very few devices to eventually
a three hundred millimeter wafer and

they have tried to do a foreigner in
fifty millimeter wafer and you go.

Bigger package more devices and
you only need one tool for it.

It's a really lean model is you have
as small of batches as you can so

that you're controlling everything
as it goes and you have you know

it's ten a single device that
then gets integrated into a phone

directly on the same production line and
you're controlling each of these so

you have a very small factory footprint
you're and you're sticking to

kind of lean manufacturing ideology and
you're doing this by constant feedback and

feedforward in the process I
mean there is this issue that

you know there are many things that
lead to this rocks a lot of problems and

one of them is if a way for goes bad
it's a three hundred millimeter wave for

the whole wafer goes bad if it's
just one chip you have much

less of that problem one chip is
probably unrealistic at this point but

you know we're you can with a company for
instance that makes a very small.

That makes a very small F.M.R.I.
and what we would like to do is have

them create a factory that
makes the chipset that and

everything else that goes into this in a
space that would normally be required for

each one of those steps a much larger
facility that's working for a lot of

different companies so we have some really
interesting examples of how this works.

And this this has some I
think geographical and

cultural implications you can have
entire factories around the country and

around the world that make entire
products it's also wonderful for

entrepreneurs or people with a I do as
they can reach out to entrepreneurs

we work with a company that is
right now getting an enormous

amount of attention for potentially
being the first quantum computer.

And they're competing against I.B.M.
it's a start up company.

In a fair hell how large is that we get
a fab do you know you've been there.

Not much larger than this room you
can imagine what their competitors

that are we're trying to work for the most
advanced Maybe world changing thing ever

which is quantum computers and
they're they're not inhibited by the find

the financing because they're trying
to create the process in such a way.

And I mean MIT just did an article just
two days ago saying that they're just as

likely to succeed at this in the next
few years and have a proof of concept.

Before the end of the year as any of
the other companies that are competing for

this all of which are more time
Altay billion dollar companies.

So you saw this picture earlier it's
reinforced this is this is what

we're talking about that
no matter how much data

you're collecting it doesn't you know the
probably shouldn't even be the white lines

in between it's a fully
connected process where

you don't consider a single
production tool or a single sensor

in a production line to exist by itself
using as you see it as a whole process and

once you see it for a single factory
that makes one thing you can see it for

a single factory that makes an entire.

You know and in and in tire not
semiconductor devices but an entire

product that would be a consumer product
or a commercial product in some way and

you know we've seen pieces of this working
such as they were getting example also

a big hard drive manufacturer I mean the
so there are a number where it's starting.

But this is the goal we place as P.C.
with A I P C It's only

now becoming possible and
I did John will go into some applications.

Of where at least on the imaging and

recognition side the feedback loop
changes from the way that it did before.

So you've heard about imaging I've barely
touched on how AI works I've barely

touched on the way Super resolution
works or on the way rapid three D.

inspection works touched a bit on
how large visions of the future but

since I've touched so little really in
the end please let me know what you're

interested in and we'll talk more about
it do you have any questions at all or

just areas that I can go into
more detail about yes please.

Yeah.

Sure so.

Well the things that I think that

the first place you can
start to think about it is.

A bit the just take one
part of a process so

we're talking with a whole factory let's
start with one part of a process and.

A microscope that uses a high.

So.

If you're dealing with
a reactor something that is

doing a deposition of a thin film on
a compound semi-conductor doesn't

matter if you have any background in this
particular area or not it's a reactor it's

a certain temperature it's growing a seed
of a crystal on another crystal and

your you you have something that
is a device that is better or

different than anything
if it were not to pose.

Now on the substrate that

happens before you put this deposition on.

There can be limits so
there can be defects on the substrate.

Those defects can be
predefined as either device

breakers that just things will not work
you have to throw it out they could

be defined as well this is a problem
that we have to look at later or

that we don't use that
section of the way for.

But you may be you may inspect that and
inspect it in an S.P.C.

way so you're looking and you're
saying there's something going on so

then you're doing adding something to
that with fairly limited knowledge but

knowing that it passes
your specifications so

somehow you're not going to
throw away this way for the.

Substrate made and so you're going to then
add something called an ep a taxi to it

and you're going to do this in
a reactor with that has deposition so

then you get through with this and then
you measure it again with a microscope.

Microscope being this AI enabled
super rez microscope and

you find that there are more defects.

Now.

Where did this happen and
why did this happen.

You could at that point choose
to throw away the entire thing

but that could be hundred thousand dollars
you know that you're wasting right there

I mean we have a customer that has an
eight inch wafer with one hundred thousand

devices on it and it costs one hundred
thousand dollars in the end because isn't

that the device so this could be an
extremely expensive mistake so you have to

know why did this occur so the S.P.C.
answer would be well you're looking and

if the if the substrate was in control and
the reactor was set

to a certain temperature certain humidity
all the conditions that it was set to.

Then we're just going to keep optimizing
it intil there's no problem in the end

instead we try to see and John to go
through this is what are the what

are the trends that are happening so if
there are hundreds of thousands of defects

in the end what's causing those defects
in the reactor so there are sensors on

the reactor that are being measured you're
taking that data you're taking the data

from the information that we've acquired
from the substrate measurements and

then the information from the at
the taxi microscopy measurement and

through that you create one final control
chart and that final control chart will

say OK Something went wrong at one place
now if you have a feedback system.

That has you know that is you know
constantly giving information back and

forth it can say actually this happened
it cracked the lattice structure

by being by something being
altered in the reactor so

it should be able to set the reactor in
real time to be able to adjust for that.

So suddenly you don't personally know

a human doesn't know
what caused the problem.

But an AI is learning through
a process that there is a problem and

what conditions change that problem so

that it can be changed in real time so
suddenly you have.

A really complex device like you say
it's a silicon substrate maybe it's

a gallium nitride attack see
that you're growing on it and

somehow because of this feedback and
feedforward you're creating it in

product that works eventually
because it's learning the way.

A simple set a human would learn you
don't exactly know why you learned

to do something but

through exposure you have learned it and
you've changed your behavior.

This small part of it produces
a compound semi-conductor.

Now the AI that does that John we'll
talk a bit about how that works but

it's this really interesting
thing that we are the data

that is pollution isn't pollution if you
consider that the total amount of outcomes

that you don't want to have for each of
these layers is actually fairly small.

So you know there are say thirty
different potential types of defects and

on both of those layers but there may
be hundreds of thousands of them so

a human can't see them they
can't do anything with them

they're too small there's
too many of them.

But their limit.

So it's what we would consider sparse
data so most everything except for

those defects you can just ignore so
you don't have the data pollution anymore

you're working within a fairly small set
and it starts that you start by actually

training this you know you're teaching
a system what a defect looks like and

because it's smarter than
we are it will get it right

it will get rid of the bad ones and
you can monitor this through time and

John will show you this of how it gets
better and when it gets better but

then the next time you do it it's
already learned and keep learning.

And so
we call That's a prior that it has then so

it gets better with time because of this
you have one small layer of this now

that exists we can do that now
if you take that same idea and

then say we're going to
print devices on it.

Well maybe you go through this entire
thing you see the process is not just

a substrate with an epitaph but
it's a substrate a taxi and an L.C.D..

That comes out of it in the end so
what process goes into that so

it just becomes another layer
in of inputs in a neural net.

So this can continually scale within

a process until you have
something enormously grand now.

We're nowhere near doing this for cities.

Entire factories that are making
from the start of a car tire to

the electronics in a car but we're doing
it all of those individual places so

that we're setting the groundwork and

the ideology to bring those together and
it's the exact same ideology.

Now if your question is So
does that make sense first of all.

This.

This makes sense OK.

You know.

Anything.

This is a system.

To this day.

So.

Give you.

System.

So my guess is that OK so
there's nobody that's doing it so

there's nobody that's doing this
computational part of it because

there are technologies that if so
it wouldn't be the most efficient there's

two parts of what we do so there's the
fact that we can make a in an inexpensive

tool that gets high resolution and
can give information that makes

factories cheaper more efficient and so
on and it's one tool instead of many but

you can imagine something that has many
many tools there's a lot of physics and

great advances and
everything from microscopy to sensors and

all of this exist so
the ability to acquire a lot of data and

to exists now what but to but

the idea ideology even of assigning
this type of feedback loop feedback and

feedforward system with inputs some
hidden layers in a neural net and

a desired single desired in
control output nobody's doing.

And to a large extent you know
we're only doing tiny bits of it

right now but you know but
what you know we will

I often think that we don't have a lot
of competitors and nano tonics so

I don't mind talking about anything we
have a lot of potential collaborators.

If somebody is minute we're going to
need the people to make the reactors to.

Able to pry the right information and
to be able to adjust it we're going to

need people that have a lot do
electrical testing and protesting and

all of these things we're going to need
other microscope makers products so

we think of this space is one
that we need to talk about so

that everybody is involved with it and
understands that it's not just your Amazon

echo they can have a system of
learning but also a factory can but

nobody is doing it and to the extent that
everybody starts if you're in a deficit

of defense of room of business
leaders the largest companies in any

of these spaces they will mediately say
we've been doing in machine learning and

collecting big data since one nine
hundred eighty and I'm sure you're doing

a great job of it you are accumulating it
you've made great progress but no matter.

Whether you've had done this it wasn't
possible it's only now becoming

possible to be this fast and
run a real time neural net so

there's nothing to do with the genius is
that maybe in those companies has to do

with you know that these technologies
have not been able to converge before so

that's why nobody's doing it maybe
it's partially a lack of vision

partially status quo thinking and
partially just something new.

Yes please.

This is your.

Right so.

Right so we we consider that.

If if I were to look at all of these
other nodes and all of the testing nodes

in in I'm in a facility
we happen to think that.

That imaging looking at
either morphology or

cow in looking at all of the pixels
that you can gather to be a first

most important step right now so
that's why we provide this

so we start is imaging I view that
at our company research were damaged

imaging a nano trying to automations are
one company but this is why were we think

of it in this way so
we do provide these bays of imaging.

And we provide a way in which we can train

a system so we have you can think of our
company as being a large a huge codebase.

That has a super easy to use front end for

people to train things now it's
the training is completely agnostic

to whether it's our tools or other tools
so we sell software for doing as C.M.

in T.M. analysis we.

We would you know we
are an imaging company and

we find that sometimes morphology can
relate to electrical properties in ways

that people didn't know it could before so
we do supply the whole system and

we supply the software separately and
we supply a service

that is not doing testing for others but
the service being we're constantly

improving your system not even this to
your physical system we're improving your

software system because we have a learning
system that as we improve you approve so

we get so honest a business sense
of it we sell physical instruments

we sell a long with the software or
we sell the software separately or

if we sell the software with the physical
units we have an ongoing a new Woody

which is being you know
the updating of that software and

keeping everybody relevant all the time so
that the specs you look at are not fair.

And you're not using today's equivalent
of a floppy disk later on let's

have the business model.

Please I think so.

Yeah.

Me.

Right.

So.

John and John will go into this a bit more
but and training we have to first realize

that the I mean the self driving car is
an excellent example of an AI P.C. at work

if you have a successful self driving
car it's a self-contained system that's

reacting to a lot of external inputs
producing an output of not crashing and

getting places as as quickly
inefficiency as fission is possible so

I P C exists in the sense of
a self driving car not the inputs

that you're giving in in our case are.

The spar system out of human inputs
possible this is the goal of what

and I'm by the way I'm not saying that we
have achieved everything this step we have

a little bit and you know I don't I
don't want to say that we are these but.

The equivalent of self driving car we kind
of achieve So what what we're doing is

to say we have five thousand or
just say a thousand for

instance cracks that have
an angle to the crack and

the length of the crack a human
would be measuring though so

you know a crack is bad you know an angle
that is over a certain degree is bad.

So you train a system you say now that
looks long I'm going to circle it

chain a system that looks like the wrong
angle I'm going to circle it you do this.

Thirty times a few of those you're just
going to get wrong you know the difference

in six micron difference you know
a human just going to get it wrong.

And the first step to an AI is to say.

Well you know we're going to get.

Rid of the you know we're going
to say that that doesn't seem to

work throughout the whole system so maybe
you should look at this and retrain it so

then more retrain it and the next level
of unsupervised learning will to be

to retrain it now when you get enough data
a human doesn't have to do anything so

the system then unsupervised understands
that you don't want these cracks and

you've run so many tests and you've done
so many things and it's fed back to this

final process of success that the car
not crashing is every chip working.

And process always being in control so
always not not only not getting in a crash

but always getting there most
efficiently Once you have enough data

humans are training as sparsely
as possible that's our big goal

in order to get to a point where
a machine continues to take that data and

put it into a larger training set so it
starts to creating its own training set.

It's almost exactly the way that oneself
driving cars are fully optimized.

They will work.

You know it won't stop
driving cars will do

is I think probably more
similar to what and

a few food you know deep minded division
of Google beat the world champion and

go the hardest most complex game that
most theoreticians says couldn't win for

you know it will never be able a computer
will never be able to win it when it won

by something called reinforcement learning
where you make a huge study millions and

millions of things and you go back you
make mistakes and eventually the AI

makes decisions that
a human wouldn't make and

becomes superior and human but
you know goal player will set.

Possibly thought of that the best just
the just go players even looking back and

think of it now we're lucky in a factory
in especially so even a complicated

thing is not as complicated and
driving a car is not nearly as complicated

as go there are there is a sparse
amount of things that can happen so

we can use other systems of machine
learning that don't require millions and

millions and millions of datasets
the way that learning Go does and

that's our big challenge is to be able
to use the sparse amount of data to get

that type of improvement that makes
decisions that human could never make and

you know in some cases we succeed in
a lot of cases we have a long way to go.

And there is another
question here I'm sorry.

Yeah John we have it with
three case studies or so.

OK Yes So we're going to a feel and

then if you want to talk about more
I mean we're lucky that we have

I get with the greatest thing is that
we're not a semiconductor company or

we're not a nano tech company we
get to see what what everybody else

like you guys are doing in many different
fields so a lot of different case studies

so I don't know whether I don't know if
you have this one but we have worked on

self driving cars we've worked you
know these thin film we've worked on.

You know.

You know all sorts of that you know
you'll get into some of these.

Later But say I said enough if you're
interested in another one just ask

me about it afterwards I asked John
because we've probably seen a lot of them.

How are we on time.

OK this is it.

OK.

Anything else before.

For this or do we want to know you please.

Yeah.

Yeah.

Yeah.

Right I mean we're we're interested
in giving you the power to train and

do that for
no matter what you're using now.

This is something we've done with a few
partners so if we have a seventy five.

Instruments in the field
there aren't a lot of that

that we have out there right now
that do that but we have a customer

that makes the next generation
battery technology that uses a C N T.

That.

Takes our software does the exact same
thing and it's kind of invisible to us.

So yes now what we don't
do is you don't offload it

this is not part of the business model
right now you don't offload data to us and

we for a price give you data back.

We put the the can we have the software
system that you would own or

be a part of a company or system or
a lab from which you can train and

work with and use the base the algorithms
that we have built into the system already

but it's no different to us is that
a good way of saying I'm trying to think.

Right.

Yeah.

Yeah I mean so ultimately.

We're interested in factories.

Because because I am interested in seeing

somebody from an extremely
large pharma company.

My board level mazing scientists said
to me and they're spending money on

a buying a bunch of biotech companies and
said you know really.

Everything the vast majority of things

have have been solved for
what we're dreaming of in chemistry

we don't they're not going to
be enormous chemistry advances.

In the scaling of that and making getting
those into the world is going to require

combinations of that chemistry that we
don't yet know and making sure that this

works in a will when they said that this
is exactly why we started the company.

And it makes as a chemist it
allows you to be more creative and

think that I don't necessarily need to you
know yes I can invent a new compound but

how do those compounds work together so
that it actually can make a drug or

can you know make a new product
that's where we fit in is how does

that point the right after research and
right at the point of production and

how do we figure out how to
get into production and so

are most interested in is a business I'm
interested in research into research but

that's so not who we are as a business.

OK.

So just.

Thank you.