This is really what we will be
talking about exactly how to make.

What the enemy period.

Maybe.

Our limitations are looking on to kill.

Them special especially academic once.

You try to share some often
experience with all of us.

Such that we can make useful
insights from better so.

Many fact he actually has a video link.

And here to himself so
we're getting this Ph D.

in 2000 suffering from MIT
electrically hearing and

computer sciences I think this is where
the seminar professors Demitra me I

decided you might so do his credit he
holds the record of the high school B.B..

All will be in their
millions on the stairs.

Which i was publishing I.D.M. 2012 I'm
using the best power paper I want to be

electronic devices letters
into present serving

in addition he has a jar you see me
now every group of them prime targets.

I work for the best master business and

very quietly the Wall Street
police that's why I have a mighty.

Way Eves working on commercial
they fell off field and

meters that have actually been
fabricated and might be criminal.

He is also in stock where D.M..

Rebirth for our research laboratory

results every 3 months or
in the same week I've read the St.

Joe's he has I think as I feared
I was a box of them for years and

this fellow see and
I must there's a loser.

So that.

Or it thank you.

I know you guys have spent a lot of time.

Inviting me here so I appreciate it and
before I go into the bulk of my talk

is do a little bit of advertising for
for these fellows down here so

it's well see is a small company and

what they do is they do fabrication for
hire and

we make greetings for
we specialize in greeting so

nanoscale solutions is my own
creation it is actually fictitious.

We began invest we began
trying to make this work as

a group that would fab for people at
the MIT clean room and or MIT and

Harvard clean rooms rather And
we just we were still looking so.

Just the just to give a little
bit of free advertising so

today I really want I want wanted
to talk about decision making for

experiments and the reason why I want to
talk about this is I feel this is a really

under-served topic no this is
a very important skill that.

That you should have and
it's really separates the good engineers

the really good ones who
are capable of doing research and

the ones who are just engineers and

I'm using the mosfet as a case study but
really I could have used any topic so

before I get into this the I'd like to

kind of find out what the audience what
you guys want to do in the audience so

just by showing hands how many
of you want to go into research.

All right so like.

Quite a few of you I see Professor
Connerly wants to do research here

how about the other half do what do you

who once OK what's the sanity check
who doesn't want to do research.

All right so really that doesn't
add up to one select try again so

who wants to do research All right so the
majority of you guys want to do research

and the message that I have for
you is you have to understand what

you need to provide to people in
order to get a job in research OK.

So here's what I found that most Ph D.'s
aren't able to get a research job or

at least the research job that they want
to get some of them want to get faculty

positions some of them want to get
research positions in industry and.

My my main message is the reason
why you can't do that is because

you don't understand who your customers
are and what they want and what you need

to tell them to convince them that you're
the right person for the position so

there's a bunch of people competing for
these jobs and they're looking for

something very specific and
if you have it they'll want you and

if they if not they want OK and the role
in academia is in general a professor

your customer is essentially
the university or funding agencies and

companies who basically whoever gives you
money right and what they care about is

that they want something they want some
tangible results from They're never going

to they're not going to give you money
unless they think you're the best person

for for the job right they're not going to
say that guy is a really swell fellow and

but he's a sub optimal engineer
let's just give them a lot of money.

And in industry.

You.

It's the same thing as a research engineer
what they care about is they want to hire

you and they want you to do R. and D.
and make a product out of what you do so

how do I trust that this person
has what it takes in order

to take something that's supposed to
fail like research everyone thinks

you know majority of research doesn't
work right to say how do I make

someone have good risk reward bets so
doing research and

working on interesting problems really
isn't good enough to get your research

shot me really have to have more skills
than that and I'll break this down.

Using somewhat of
an industry hierarchy OK.

So the required skills are kind
of they are on my right and

the roles are on the left and this
pyramid with the number of people kind of

shows how plentiful this role is OK so
at the bottom there's a lot of

module quality yield engineers there's
a few people doing integration device

type work at major companies these
are kind of you know development job and

then really at the top there's R. and D.
and pathfinding which is where people

really want to be when they say I want to
do research right and the skill set is

larger for the R. and D.
and pathfinding people than they are for

kind of development engineers so
at the very bottom.

You know the majority of your engineers
they all have it OK I have this problem.

And I want you to fix it once you tell
them what you want them what you want them

to fix they're very good at doing it OK
but if you want to move a step up and

go on to an integration engineer great
you're going to be in charge of a few

hundreds you know 100 steps something like
that if you're at a very large company

you kind of have to understand to some
extent what those 100 steps work or

do which is why you need
partial flow knowledge and

you have to be able to manage
projects a larger projects OK And

then when you're talking about our Indian
past finding people these are people who

you're who are going to be making
the direction of your company OK So

they're they're going to say I think
you should go in this direction and

you really have to know
some risk management and

be opinionated in order to like have
this opinion or to choose a direction so

that's those are kind of in blue those
are skills that people generally it's

more of a personality trait rather than
you know something that can be taught but

here I'm trying to teach it today so
this is why I think this is important and

I want to talk about how you
make decisions in research and

it really is a simple
strategy I could probably

summarize this in one sentence but
I'm not going to.

So the goal really so

I'm going to use a case study just
to just to prove my point and.

I see this a lot you're essentially a new
grad student in grad school year you know.

Perhaps your goal is to
make sure channel mosfet

in your group has never done this before.

All right but your previous student has
made this long channel mosfet before right

and then you're you just your goal is to
make the short channel mosfet and as a new

grad student you're like well I've read
some papers and things like that and

I'll just adapt this process this whole
process and meet the new goal because

like why would I bother reproducing this
because I can't get a paper out of it or

whatever right so you just really
want to get to the end goal and

you want to achieve something OK and what
happens when you go straight to your and

well you get this device it's like
you know a crappy device you have

maybe 2 orders on off
it's history you know

substantial swing looks bad things like
that it's just a bad mosfet right.

By the way this is not actually
real sort channeled out I just

picked it out of my other data.

So what do you do after
that well as a grad student

as a 1st time grad student you're always
like hey there's something wrong right

I made this mistake during this experiment
and you kind of reiterate on this and

then at some point you figure out well or
your advisor says OK well let's take

a step back and let's make sure that you
can reproduce the previous data 1st and

since you've never done it before you
kind of get run into this case of OK

well you know I made this moss
that it doesn't look great and

now you're back to square one you're
essentially trying to go back and

reproduce what this person did
you can reiterate on it but

there's a proper way to do this OK And
really the proper way to do

this is kind of to do process development
for this long channel mosque and

do technology verification All right so
how do you.

Companies generally do technology
verification in a step by step process OK

you take your end goal and then you break
it down into specific milestones and

that's that's what I'll show you and
how it looks OK so

here you are in your starting position and
this is just the schematic of a mosque

this is you know a very standard
long channel mosfet OK and

all mosfet is a 3 components
it's a pin junction so

this junction right here from
the drain to the body OK and

there's contacts so you have the contact
resistance the drains of the source and

you have the spreading resistance and
then you have the gate oxide which is

a capacitor a here and those
are the 3 components of the mosfet So

if you break them down you can you
can run very short experiments.

You can run very short experiments rapidly
and then you can kind of put together

put them together and hopefully it'll work
if you've done the design of experiment

correct so why do I need to make decisions

if it's this easy well you have to make
decisions when things go wrong and.

I believe very firmly in Murphy's Law If
there's something that can go wrong

it will go definitely go wrong in your
experiment and you have to plan for

things fail OK so how do I go about
making a decision when things go wrong.

Right the 1st thing is what you want to
do is you want to design an experiment

to understand the extent of a problem all
right and if you have a single module like

a pin Junction this is very easy to
do you just build a pin junction and

you put some test ruptures here in
there all right and you want to what

you want to do is you want to fill
out this chart of pros and cons OK.

So the pros generally the reason why
you're doing an approach is because you're

you know it's good this is why
you're doing it all right.

And you run into a difficult
problem now the next

thing you want to do is you want to list
out all the other potential approaches and

in this case I've just put
some really generic things.

And approach being on approach see so
the approach being

is just well there's a higher performance
lower thing that you can do and

then there's this you know lower
performance faster thing that you can do

generally always there are right and

then after that you pick an approach
based upon the pros and cons OK So

ideally like to fix your own
problem all right so what's to

let's take a look at this from a real
case this case is coming from.

When I was developing a pin junction or

you know the solar cell module for
MIT fab cores OK and

what they said is Winston We need
a we need a simple in time efficient.

Experiment so that we can show our
students how to make a solar cell

right and we need to be able to
do it in 2 lab sessions OK So

that's 6 hours all right and
we wanted to be good.

So the 1st thing that I did was I decided
OK Well solar cell people have done this

before you just deposit a poly silicon
layer on top and then and plus poly

silicon you do some thermal oxidation you
put on your anti reflection coating and

we'll call it a day I chose this
because it's a very simple process but

when you look at the device
characteristics here you kind of see well

you know I have a lot of leakage all right
and that would not make a good diet so

will just show you a is
the really easy example

because everything is known we already
I mean all the technology is there so

there's not really
anything to talk about so

the 1st is just to design an experiment so
I make a diet and then

what could go wrong with the diet Well the
doping concentrations not high enough so

you get a leaky diet right you get high
generation in the depleted region that

can come from contamination or defects or
the depletion region is too small and

you have high electric field and
then therefore it leaks really hard.

So what you can do is you can just
measure the scaling of the leakage kind

of confirms hey this is if it's scales
of area it's this is a bulk leakage.

And then you check
the contact resistance and

you see well the contact resistance
isn't great so the leakage so

the doping concentration isn't good and
finally you're unlikely to

get contamination in our CMOS lines so
you can kind of just roll this out and.

And if you just do the 2nd step here
you just lay out all the pros and

cons you can see that you know you have a
lot of approaches and really only say 3 of

them fulfill the requirements of it being
quick and you kind of can choose from that

well this is quick this is I guess I
know implantation is kind of quick and

epitaphs he is quick but we kind of have
this in-house so you know my decisions

actually really easy one of the things
that I want to note in research is

a lot of engineers don't know they have
this option it's the do nothing option.

So the do nothing option is to call
it quits and kill the project.

And actually it's really important and
I don't think people do this enough but

in this case you know I kind of can't
because we have to make a lab for

the soul for the class so

really this is pretty straightforward and
at the end what would have happened

was we just we have our own in-house up
a taxi we just stock up on the wafers you

can just grow 100 of them whenever you
run the class you just pops them in.

And these are some pretty decent
solar cells there are 13 percent and

you know to make it in
2 hours is not not bad.

Now let's move to a more complicated
situation how does that change from

before Well 1st of all.

Well 1st I'd like to highlight that
the decision making process for all for

experiments is generally all the same and.

What's different about a more complicated
situation is that the full information

isn't available you don't know
how bad these cons are and

you don't know how good
your pros are right.

And all your decisions are linked
if I want to change one thing

it happens to change
other things OK this is.

Somewhat common sense and the most
important part thing that you should

remember is that it's really important
to find the showstopper early.

So you need to find what

if you can identify what screws you
in your process that's really what.

That then you then you've
gone a long way and

you also have to keep in mind that other
people are trying to solve this problem

you can kind of watch what they
are doing and just learn off of them so

that's that's another
idea that you can use.

And I'm sorry I might skip through some of
these slides a little quickly I'll go but

if people want to go back to them
this talk is a little long or

it and I want to talk about I want to use
the case study of hike a metal gate OK.

The case for it is on the right
essentially at some point in the silicon

industry realized hey you know we
have too much gate leakage and

our devices are leaking too much they're
burning too much power All right so

the solution to that was already somewhat
known you can use this material called

High K. to get the same capacitance
with much lower leakage is OK.

And they wanted to implement this in
industry but it's heart all right.

So here's the here's kind of
the outlook of the mosque

in blue these are the things that I didn't
talk about earlier in a single device long

channel mosfet All right so
you kind of have to think about function

in order to make a circuit you need to
get your threshold voltages correct and

you need low leakage highly constrained
high performance transistors

there's a lot of different things that's
required in a in an integrated circuit.

On top of that you have the all 3
of these things link together so

your gate process is linked with your P.N.

junction and contact processes because of
thermal effects OK if you grow that so

thermal silicon oxide OK that
impacts your P. and junction and.

If you're doing your PIN
junction after your gate offside

that's going to be your gate
outside especially if it's high K.

it's certain to start mixing together and
it's not good for your device right.

And then so
this is how history goes the in 2007

Intel publishes their I.D.M.
paper revealing the $45.00 nanometer node.

That they're implementing Hi Kay for
the 1st time all right and

no foundry is has high K. right now and
everyone's this is if you're a foundry.

You know you're in for
a really long 2 years because

it's like OK we've got a catch and you
know that it's possible now the races are.

There's really.

2 or 3 ways depending upon how you look at
it of doing hike a metal gate All right

the and really there's only
the 2 difference in step so

you deposit you do your isolation you
deposit your hike a metal gate or

a you deposit your guess your date
material you do you do I know implant for

a short strains and
then you go through standard back and

process to finish off your device and when
you're doing a gate last process sorry

this is called date 1st and then this
is called Last done here you instead of

having used deposit your date but
this is actually a dummy OK so

you're going to take that dummy gate
it's just going to have the shape so

you can align your eye an implant and
then you're going to get out and

then redeposit your gate later on so
this is called a gate last process and

both of these processes are self aligned
this one obviously being the bottom

obviously being the more
complicated of the 2.

So here's 3 different T M's You
have 3 choices you can deposit

your guy dielectric 1st or last and
you can deposit your electrode 1st or

last and everyone in the industry
at this time knew what the pros and

cons were right the benefit is
everyone had been doing this

1st process for silicon oxide for
thermal silicon offside and

the benefit is hey we're already doing
it it's going to be smaller than

you know this gate last process no one
has experience with this right but

the problem of that is because you're
doing your You're source trained

contacts after your gait you're going to
run into these thermo budget problems and

then when your medals and high K. are
mixing together the you're going to start

running into like mobility
reliability problems tuning so

threshold voltage tuning is just going to
be really challenging to do all right but

equally challenging is are these 2
processes that nobody in industry had ever

done well they have but only in the 1970
S. had been doing and these are these

great last processes and but what you
benefit really heavily from is that you're

decoupling your thermal budget from
your gate oxide to your source train and

in doing so you can make higher stream
devices which would be higher performing

or right and the difficult
the difficulty is complexity in cost and

also design rules so
they started implementing some really

strong design rules in or like you can put
your gates perpendicular to each other

they all have to everything has to
be directed in the same direction

you know stuff has to be parallel lines
and obviously the last the dielectric last

process is the lowest
thermal budget on your gate.

But it's also the most sophisticated so
what Intel.

All right so what Intel
implemented the 1st year was this

dielectric 1st electrode last process.

All right so everyone knows what
the positives and negatives of this was.

But people just don't understand what
they didn't quantify they don't know

to what extent is how good how

much something hurts you or
how good something straight and

as the new as new information appears
a winner will definitely appear and

there will be a you know a right way
to do it so so let's take a look

at history before I do that I will talk
about information gathering strategies

at this point because this is this is
what's going to be really important.

So I have really boiled it down into
3 strategies 2 of them are generally

used in semiconductors The 2nd is not so
the 1st one is the shock and

method everyone is familiar with
the shotgun method you kind of if you say

somebody has no direction that's because
you're using the shotgun method and

trying to go in all directions straight
it is the try everything approach and

the benefit of it is that
you're trying everything and

you don't get blindsided but the downside
of this is that it's really inefficient

on your resources because everyone's
going into you know direction so

why don't you use this you use
this when you really don't know.

You when you really don't know anything so
if you look back here

people just don't know what
which one is the worst and

which ones the best right so you kind of
when they'd started with the past finding

in this they tried everything and
as you get more information they can

kind of gather an opinion and
bet on something OK I think

this is actually overused in research so
that's that's kind of my opinion.

The 2nd is bootstrapping you really
don't see this very often and

semiconductors just generally because
there's not a good way to not bet a whole

lot of things a lot of money on something
so you really what this is doing is you're

betting a small amount of money on a hunch
like a well I mean people use it for

hey could you run a few experiments on
this I'd like to write a proposal right

it's a way to cheaply does the risk
in approach but the problem is that

because you have so little resources
it's slow All right so but

once you have enough information to form
an opinion then you generally move on.

And a lot of people use this for
startups just because

it's a cheap way to to
get some information.

Just an idea what can I do other than
that while I 1st shock on method there's

a lot of other things that you can
do nowadays generally you are not.

You're not doing something brand new so
you don't have to try

everything experimentally you want to you
can read some trusted literature you can

hire a consultant who already knows what
they're doing and save you a lot of money.

More more advertising right.

And or you can use an already
accepted approach in just go with it.

And for bootstrapping you can kind of use
sanity checks to ensure that the idea is

worthy of dedicating more resources and
be more aggressive better so.

But some of these things you
can't really do that and

you really have to prove
it out experimentally.

So the point is use your
other resources when you can.

The last is kind of the holy
grail you want to be

you want to end up making directional
bets what a directional bet is is

you take you're basically
almost going all in and

when you do that you get a lot
of progress really quickly OK.

So when you've when you pump all your
money behind doing a development for

a certain for you know for

one of these technologies you're going
to be able to move it a little into

maturity very quickly you know what
we know foundry bets on all 3 and

releases 3 different process no I mean
it's just way too expensive to do that.

But the problem with that is that if
you're wrong you've lost you've got all

your resources on this one path and you
lose you're going to lose a lot of money.

So you only do this when the answer
is close to being known and

this is when risk management comes really
heavily into play if you don't manage risk

when you make a directional bet you're
you're not going to have a good time.

People rarely use this during research and

I think that this is
because they have FOX OK so

what is fogs Well everyone's heard of
foam It's the fear of missing out and

I've actually only heard this term
recently and fog apparently is the fear

of getting screwed so I think that this
is what this is why people don't use it.

So yes this is the holy grail.

What I would like to note once again
when you're making a directional bet

you actually need to know exactly what
you want to see when you're correct

OK this is how you confirm that hey I did
this correctly and you need to know what

the signs are if you're incorrect and
if you're incorrect you need to be

you need to be very sensitive to the facts
pointing that you're that you're

wrong you can't cherry pick your data
you can't be like well you know for

my short channel fat Yeah I just made
this mistake in the fad if I do it again

it will work this time those are those are
things that you don't want to be doing.

And really you have to be willing to
cut your losses at strong indications

that your hypothesis is incorrect
otherwise no matter how much effort or

money of you've invested into it otherwise
you'll be end up doing what this

guy is doing which is pouring money
into the ocean in a sinking ship so

this is these are the things that are
really important and you really have to

know what you're doing when you're making
directional bets or back to the example.

5 major players in
the semiconductor industry Intel

was the very 1st one they did High K.
1st metal last in their 45 generation

now this next generation
everyone else joins the party so

I'll spread this out and
you can kind of see that everyone has

very divergent approaches everyone's
tried something different OK So Intel.

Transitions to hike a metal gate
last technology OK yes some seed

does hike a 1st metal gate last and
then Samsung GLOBALFOUNDRIES and

I.B.M. part of the have an alliance so
they're all using the same technology gate

1st technology although Samsung's
looks a little different.

Take the next generation and
this is Intel moves to Fin FET but

they're still using high K. metal gate
last and some C. and Samsung follow

what Intel did last generation and
they moved to hike a metal gate last

while I.B.M. and Global Foundries
continues on their gate 1st approach.

So this is during a time when
global isn't doing very well and

to some see instead of some very smartly
said well this is what Intel did

they're on their 2nd revision
this is probably what we should

THEY Intel is really I
want to stress this intel

is the leader in semiconductors and
they're generally the 1st to do it I hear

a lot of people talking poorly of Intel
but really they're the interveners here.

Finally if you look at the 14
in 16 on a meter nodes

there's everyone converges back
on to hike a metal gate last and

Fin FET technology so really who
got it right well these are some so

you probably got it right 1st well
Intel got it right 1st sorry and

then test some C. followed and
Samsung is in 3rd in the global and

I.B.M. are dead last I didn't
plot the next generation.

You have these global and I.B.M.

are out of it so you can kind of see
the ramifications already happening.

And the results of this case study
is really hey you know to some C.

comes from not even in the top
25 you can't even find

them there to number 3 and
the semiconductor by revenue is not just

the fact that they got it right there's
other reasons but this is the results

this is why part of the reason why
they've dominated so hard because they

you know just one node made a difference
you know kind of made a difference and

if you want to quantify this this
is a factor of 10 growth in revenue

now lastly I'd like to talk about
the differences between so the last

example that I used is you know that
the end result is possible right now Will.

We do research OK We do research and
what it means is that.

We don't know if it's physically possible
that paper that you read that ran

the simulation OK that's a hey you're
going to get a Terra Hertz rate

that may be made that may be wrong and
you don't know at the outset whether or

not it's wrong or not so
they may have left out key physics and

when you leave out key physics
your simulations look really good.

The other thing is that you can be it
could be technologically impossible

right now we just don't have the
technology to make perfect materials so

you know right now this is impossible and
if I mean here in the long haul for

10 years yeah you know
perhaps it'll work right but

you has a grad student may not
be one to be in the long haul.

So the question in development
is really is it economical

I know that this is
possible how much money and

resources do I have to pour into it
to get this to work and research.

The research is kind of Is this
feasible can I get this to work and

the way that you think that you go about
it is going to be different you're going

to require feasibility studies
your resources are less and

you're kind of experimenting to prove
that hey this idea is possible or it.

So I'll actually talk about my own work I
worked on stranger mania and I'll tell you

why stranger mania is good to begin with
so this is the I.V. characteristics from.

I appear to be doing very good on time so
there will be no skip slides.

So what you really want to do is you want
to minimize the power right now we're

scaling the mosque fats and therefore if
you want your power density to be constant

you kind of have to drop your power
All right so the main turn is this or

the supply voltage the very 1st
Ratatouille people came up with is

they said OK well we don't really
need all this off we don't need

such great leakage floors if you
just shift the curve to the left and

we can get a lower supply voltage.

But that runs out of
steam after a while and

then you in the black curve it's like
well we need to do something more so

they implemented ways of improving
mobility things like strain silicon.

Using 35 materials germanium and

that's why people invested your money and
it's for people it's all right.

So here's the pros and cons charts for
the chart for germanium or

stranger meaning rather for silicon people

it's the it's it's what's already
out there everything's mature but

the problem is that we would prefer if
we had a higher mobility material and

we would also really like it if
the native oxide is you know higher K.

rather than a $3.00 right so
that kind of kills your oxide scale.

For stranger medium you can get really
a significantly higher mobility

than the silicon but

you have these detriments that are that
you have a lower band gap that's higher

costs and the gate dielectric
technology is not quite as mature.

So at the beginning you're going
to have to use the shot well

you don't have to use a shotgun method
because you're ready have an approach

you just have to 1st of all prove
out that this is indeed true and

then quantify what they what these
detriments are and whether or

not any of these things
are showstoppers So

this is kind of what I
say if was referring to.

This is this is our proof that string
germanium indeed has high mobility

this purple line right here is
Intel's data and this is and

the rest is kind of my data so
if you look at this data over here.

Intel's quantum wealth.

Falls very well in line with my by
actually strain data and this is about

a factor of 4 improvement on strength
silicon so really we can see that yes.

STRANGER meaning has high it can have

a whole mobility is exactly
with what we would expect.

All right so we've kind of shown that hey
this is feasible but now what's necessary

for a short channel technology is that you
need to be able to scale the gateway and

what does that mean I need to be able to
scale the electric I need good contacts

I need good you know I need to see
that this device can meet the off

state characteristics that I need
at the channel links that I want.

So the 1st thing that we did and it's the
1st because it's easiest OK I took this.

And previously I do want to
mention that all those T.M.

that you saw a hike a metal gate
they they were from chip works

those guys have reverse engineered
everything there now called I.C.

insights I don't want to
skip giving them credit for

that but this work is from
this this is one of my works.

And I took these data points from
literature so you can kind of see that

as you scale the gate dielectric down here
your mobility just drops really heavily.

And this is something that's
well known in silicon and

it's important tank kind of to quantify it
because what it means is that the benefits

here on stranger medium are not as good
as what you would expect them to be so

you're kind of saying wall you know

it was really good now it's only kind of
good is this really worth the effort.

But at the end what really kills it is
the off current versus the gateway OK So

these are all the germanium devices that
have been published in recent times and

when I say find the showstopper I mean
this is the showstopper there's other

showstoppers that can
be fixed this one is.

Well it also can be fixed to is just
going to be very hard to do so so

this dotted red line right here.

So these pink dots right here artists and
scenes work and I've kind of trace this

dotted red line is kind of best you can
do in terms of the upstate current and

100 nano amps per my Chron is the high
performance leakage targets and

100 Pico from Micron are the ultra low
power leakage targets and you have

many transistors ranging in between
these 2 targets on your chip so

what you can see is hey we want to be at
maybe 10 to 20 nanometer gate length so

if you extrapolate this line yeah we could
probably meet the high performance off

state currents but what's the problem
here well the problem is we can't meet

the ultra low leakage and really
you're not going to just put in some

transistors that can meet one set of
goals it's just too expensive so.

The leakage floor is too high and that's
that's what we found to be the showstopper

in germanium So really silicon wins again.

Is just so silicon is a very
well engineered thing and

and we found the showstopper and
we'll just wait kind of for

somebody to fix it because you
know we're just not interested in

solving this type of problem people
can engineer it they could do

you know doping engineering things like
that this is just you got to be in for

the long haul you've got to really put
in some effort to get this to work.

The other sign that we're that we're on
the wrong track is really you start to see

Intel and T S M C publish papers
on germanium and it is you

if you take the hint it's like well
they publish either something that.

Product or something that they
don't think they're going to use so

you can kind of take the hint like
this is probably not going to work.

All right I finished
faster than I expected so.

Today I kind of hope that I
drilled in these 3 things to you

the decision making in experiments
really boils down to 3 steps

it's you design an experiment you to
understand the extent of each problem and

you kind of list out all
the approaches pros and cons and

kind of quantify what the benefits and
drawbacks of something is and

then after that when you want to make a
directional bet you pick an approach that

you pick an approach based upon
the pros and cons of each and.

Kind of sometimes in line with your
current research resources like hey I have

a lot of time so or have a lot of money
I can kind of take a different you

know one of these approaches and I would
say don't forget about the do nothing

option where you actually kill the project
because you don't think it's feasible

in a given your resources and
I really like to stress

a lot of the decision making process is
related to the gathering of information.

And there's an art to it
OK there's an art to.

To doing this which is design of
experiment you know the scientific method

that you're taught in 7th grade is
actually very good just nobody uses it

and.

You want to decide you want to
hone in on your show stopper and

decide whether or not to fix it and
with that I'd like to take any questions.

If you were.

You.

Know them a little bit further the up.

Friends.

Were.

Like.

Well.

So you are locked so
you have constraints and

you kind of the constraints are listed
in your you know your concept.

And you know if you're the guy
working on 10 I mean or

plus plus plus per se you kind of
you have to show like hey if you

you know tweak this like this
is this you're kind of still

in this mindset of hey if you give me this
much money to develop this process and you

tweak it in this way we'll be able to make
money off of it that's still the mindset

that you should be thinking and it kind
of doesn't change what you just have

a lot of constraints that the yeah I can't
change you know how we make the fan and

I can't change it like you
this is a Fin FET process so.

From that point of view yes so answer
quite short answer yes it's the same.

000-000-0000.

Stop by line like.

So if you don't have
the resources to like.

To branch out so generally as a as a grad
student your time with all your time and

money limited so if you're using
the shotgun method it's more or less for

design of experiment like hey I
need a deposit this dielectric and

I have no idea what the process conditions
are right you generally won't have

the money to go down or time to go down 3
different paths what ends up happening is

hey I'm going to try these 3 things and
you know I get nothing done because I

couldn't don't have enough time to fix
any problem and one specific thing and.

And when you're talking about this type
of stuff people are competing against you

OK people who's you know
people whose full time job

is just to do one of
those projects right and

you're not going to the you
might be smarter than them but

if you don't have the bandwidth to deal
with that it's just not going to happen

you just kind of let that project go
you can ask somebody else to do it or

if you really have a good hunch you can
maybe invest just a little bit to convince

your advisor hey you know let me work on
this full time I think it's a great idea.

So there are a lot of people doing it and
I'll say Cornell and

Berkeley they have consultants
working in their clean rooms there.

Very good all right and what you're
facing is kind of economics and

you have some sort of experience
there needs to be a fit with Hey I

know how to do this and
your project isn't particularly large but

also you kind of need to say well you
know how much money does this person have

can I actually do it in in this budget and
that becomes the question.

In turn so
at most in the business you're generally

going to be facing economic challenges
rather than technology challenges.

And I'm also targeting generally
targeting say like a foundry won't take

an order that's very small so if you want
to do something small You're might come

to go to a consultant to do it and
prove it out 1st or it.

Hung in then.

Go ahead.

Yeah.

It's a misquote.

But I mean.

People are.

Going to.

Look at it like yeah and.

It was just.

Doing.

What.

The.

Last Wish.

OK so.

So the 1st 1st question you're
absolutely correct emotions

will be there especially if
you're a grad student and

you're doing it on your own you don't
have you know in industry there's

technicians right kind of
say like hey run like here's

I've drawn out these steps and
year you run it right so

that kind of gets taken a little bit out
of the picture and you're absolutely

correct like yeah it's really frustrating
when something doesn't work great and.

If you're a professor you kind of have to
if you tell a grad student hey you know

do this and it doesn't work long enough
the grads eventually that will just like

give up so you're absolutely correct so
2nd could you yes.

Right so one comment about industry
industry has a huge amount of money OK So

industry has you know when you
have say 10200 people working

on something that's you're
pumping through like 204 for

2040 1000000 dollars a year OK And
I think of it in grad in terms of grad

students they're seeing $40.00 to $40.00
to $80.00 grad students per year for

full 5 year grad students of data and
just by salaries alone and

you're kind of that's that's the scale so
industry generally knows more and

with Bell Labs it works until the until

people start being wasteful if
they don't make products out of it

your research just because I mean it
just falls over at the beginning it

works really well when you have very smart
scientists at the end when everyone just

like doing random stuff your business
model falls apart so and lastly.

Yeah so.

1st Intel is definitely the leader.

10 nanometer I think I saw the process
I didn't get to dig deep into

the process last year they
released a process technology yet

I.D.M. very impressive looking they're
having I'm assuming they're having wrapped

up issues right now they did
a lot of they you know they're

they're well on their way to
continuing this history of.

Of hay were leading the industry and

I think that in toll maintain their market
share I think that you know the C.E.O.

of A.M.D.
is a former group made of ours or of mine.

And she's doing great things but
I think I mean Intel is so dominant and.

There you can't really it's
like not a comparison.

And he and any other questions.

Are it so
we'll go with you 1st because we're.

Everyone likes to I
mean everyone likes to.

Everyone likes to beat on
whoever's at the top it seems and

yes this is the industry's
no exception so.

Well is that.

The perhaps is jealousy but sure.

So stranger medium is the project
that I was given by it at the end

you know I'll quote the Lisa Sue She
was the what if she

was the speaker at the at the wedding
ceremony she said Make your own luck OK

that's one of the takeaway
messages that she said and

what it really means is that
make yourself out there when you

when when you just make yourself out there
when whenever there's an opportunity

sometimes you can volunteer and take the
opportunity just to learn things like that

and you know it kind of revolves around
my philosophy I came to grad school

not to you know be a professor I
came here to learn to write and

as as you help other people and
you know you just and

you you see other many things then you
gradually get a large knowledge base so.

That would be my recommendation to grad
students you know don't just look don't

just work on your own project forwards
your research but it also like

it also gives you only one
perspective into a specific area so

all right Anyone anyone
else OK last question.

I think I'm going to
speak blasphemy here so

I think that the area that I learned
decision making best was business

is going to take classes at your
business school and take me you know.

Because all those guys do is they make
decisions they you know weigh their

pros and cons versus each other and.

I mean a lot of engineers really hate
the fact that they work for M.B.A.'s but

I'd kind of say that I think the M.B.A.
is honestly make better decisions so

I think that could be a that that in
my experience really has been what

has helps has helped me make
good decisions or learn to

make decisions I wouldn't say my decisions
are necessarily good time will tell.

You one more thing is I plan
on giving this lecture for for

mits 5 course if you have any feedback
please give it to us or Yep please

give it to us if I wished I had my email
I was going to just put it up there but.

Yes So all right.

Thank.