Pleasure.

Good afternoon everyone is
a pleasure to be here and the bout

some of the research we've done I am
you on a chip I am you stands for

inertial measurement units and

these are mixed domain Microsystems that
actually employ a men's technology and

seem OS to create a micro
system on a chips all

the talking to a little bit about
that I am fully affiliated now

with Georgia Tech fulltime a Georgia
Tech I am putting quatrain a little here

on this page just because I'm using
some of their material during this talk.

So we're going to talk about
anybody that systems today and

when I talk about the integrated systems
these are as I mentioned multi domain or

mix still main Microsystems in which
you are using a number of sensors and

actuators that also call
transducers sometimes or

energy Harvester elements and
they can have other

they can be different in nature from
electrical devices they can be mechanical

terminal chemical magnetic and
radiative and then these devices

they require to actually have low
power alone noise electronics.

To read out the non electrical signal
that is generated from these devices and

is converted to electrical then.

This the circuits with
process that that signal

through an analog front end digitize it
through an analog to digital converter so

we called this portion of
the electronics the A.F.E.

for Analog front and then eighty C.
for back in and then is passed on into

a microcontroller to do some visual
signal processing on it now you may

think that this subsection the electronics
is actually fairly mature and a lot

like this is almost true is not completely
true when it comes to interfacing with.

With not electrically devices.

The reason is that these sensors and

actually there's produce
very very small signals.

With to just that are in the order of
micro volts currents of our peak femto

amps and capacitance changes that
are in the order of at two ferrets and

sometimes even a smaller than that so.

These micro systems are built using
the head through genius integration which.

Every For level packaging instead of the
traditional type of packaging to create

a micro system that is immune from
effects of them Vironment and

that a lot of times you need to add
functionality that are new to circuits for

calibrating and compensating these
devices in different conditions so

all together you have these sensors and

actually there's a lot of fabricated
using memes or NEMS processes and

then you have your Seema Haitien
a specific in a the circuit to build these

type of Microsystems so you can think
of many and see Moss as hearts and

brains of Microsystems today many
of these devices many sensors and

actress those are the of integrated
systems that are providing paradigm shifts

in sensing communication as well as
navigation one example again I'm showing

a picture from my start up
company is a multi-axis inertial

sensor that provides inertial navigation
and this is for use in under strict indoor

navigation for example if you are in
a mall inside a building and you want to

walk around then have the same type of
navigation capabilities that you could get

from G.P.S. but when to G.P.S. signal
is not available inside a building so

that is now given provided by these
type of inertial sensors that you see so

this is a picture of the a sick that
is processing the signals coming

out of these individual silicon chips

that are now mounted different in
the different orientation inside.

Other examples are cognitive
radio these are agile radios that

are not fixed to communicating in a set in
a certain specific band of communication

they can change their communication
band given the amount of traffic

that is actually present at any
point in time and provide a more

efficient use of the spectrum
environmental sensors can be used for air

quality monitoring the same way that you
know if you are actually using inertial

sensors in the same coffee Gratian we can
use invite and wire mental sensors so

in general seem OS processing brains
enable reconfiguration calibration and

fusion of these signals
become plex functions.

But not this talk will concentrate
on inertial sensing and

these are motion sensors motion.

Is now everywhere you look at
industrial applications for

a lot of robotic applications automotive
applications Tanika stability control

anti-skid control whenever you need
to have a measure of motion you see

motion sensors in play guidance for
drones for

F. conference role and
a lot of consumer applications including

applications in variable devices
in the Internet of Things and

also camera seventy's ation when you use
gyroscope for establishing the digital

camera pictures so.

The grand use of inertial sensors However
as I said is in personal navigation and

where these personal navigation This is
strapped down inertial navigation systems

for indoor and pedestrian application
if you look at traditionally how will

be of use gyroscopes and accelerometers
in navigation engine systems so

you know traditionally done for
navigating airplanes or missiles or ships.

These were using what is
known as gimbal system so

pretty heavy gyroscopes were suspended and

they're using inertial forces to to
determine the heading of the vessel or

the airplane and
to provide having it from.

We would like to get is the same type
of functionality but on a chip and

then integrate it into our smart phone and
consumer electronics to be able

to do the same thing so what we need
are high performance accelerometers and

gyroscopes So they are often
implemented using the memes technology.

They need to be integrated
to be commercially viable

to use in such a platform because.

The business model there is that each
one of these chips they have to go for

a maximum of a few dollars and few
meaning two to three dollars depending on

the functionality so how can you really
integrate such a complex system onto

a new chip devil of a small size and
bring the cost down to that level and

so they have to use the say
fabrication process to bring size and

cost benefits they also have to deliver
performance which means that they need to

have a large dynamic range in Bath
went to be able to be to use for

multipurpose applications two for
applications ranging from gaming for

example images to and
as well as navigation also

they need to be very robust and reliable
they need to be immune to shock and

vibration they can break down if
you you know drop your phone so

how is that coming into
play in a system of

of vedette small size so let's take
a look at inertial measurement units and

see what it is consist of so
it could consist of

a tri axle accelerometer to measure
linear acceleration to say you rock along

the line you need to know
that you're ration then you

indicated the couple of times to get the
velocity and eventually these placement.

Formation.

Then you need to have gyroscopes that
measure angle of rotation so so that

you can determine your heading if you turn
you can you should be able to determine

your heading and these devices usually put
out what is known as angular velocity so

you also need to take an integral again to
get to exact angle information now every

time you take these integrals what's going
to happen is that if there is an offset.

Added to the signal it gets integrated
over time into this integration process

and it will add to the overall error so
any type of offset that you have in

the system will be very detrimental
to the overall performance and

you're going to see me talk about that
offset in particular because this this

process if there is a little bit of loss
it is going to show up in a huge way when

you actually try to extract you're getting
your accurate position in the space.

So this information is then given to
what is known in it as an inertial

navigation system to give you position and
orientation often times you need to

add magnetometers to this set up
because when you start something

at the you know Orient even you.

Originated the process you have too
many initiated you have to know your

regional heading you need to
know if I'm heading west or

over east so at that point in time you
don't have any motion information so

you have to have this magnetometer
to provide your initial orientation.

And and
sometimes you need to have a barometer

a pressure sensor that tells
you if you're on the first.

Or the second floor of the tenth floor
of the building so all of these things

they need to be put in into a processor
that needs also a very accurate

timing element and you will see that
this timing element often times use I.Q.

crystals that can be implemented
using them as processes as well.

So I have to gather.

You know once you have the global
position added to this whole ensemble

to tell you that OK for
OK just entered the building so

at that point the global positioning
system is going to give you a fix and

once I am inside the building then
my initial measurement unit will

take care of my navigation
inside the building so

I together then this carbon thirteen
would actually give you a fusion of

all these information to predict
your you or your position.

So I mention gyroscopes being one of
the crew Well I'm an indeterminate

in the performance of this inertial
measurement unit So let's take a look

at the vibratory memes gyroscope and
see how it works these are mostly based on

twenty four cost structures and
most of you guys are familiar with.

You know basically what it is is that
if you look at a tuning fork and

you can implemented on a piece of silicon
you excited into its first resin and

modes which is score spawning the kinds
of the twenty four resonating in plane

and that you're doing that by
applying a voltage to create

a force on the body of these two twenty
fourth's and the force is then translated

into acceleration which corresponds to
a certain resonance frequencies to excite

this device into its resonance frequency
Now when you expose the device

to rotation what's going to happen
is that the rotation when it's.

Superimposed on a on a structure that has
a linear velocity would cause what is

known to Coriolis force which is
perpendicular to the plane of rotation and

the divil loss of the self which means
that the force is now in the we're talking

all direction through this
first Origin of libration and

that's going to happen is now that that
force will cause a deflection of the proof

masses indoor toggle direction and
this deflection is what needs to be sent.

So with great accuracy to determine
our exact angle of the formation or

angle of rotation.

Now most of the devices that are currently
available so you probably some of you may

know that there are gyroscopes available
in our cell phones for example today most

of those gyroscopes are operating in
a regime of operation which is known

as Mo the split and what this says is
that the drive and sense frequencies and

these are the two frequencies of
the just showed you this frequency and

this frequency are actually
different from each other.

Now when they are different these
are second order systems what's going

to happen is that assuming
that you have a high Q.

resonance if the two modes
have a difference in frequency

your transfer function is given by this
expression here which is purely based on

the ratio of the two so you need
to bring the two resonances closer

to each other to actually be able to take
advantage of the Q factor of the modes and

this Q. factor can be very very large so.

If the two modes are exactly on top of
each other then if you have a large skew

your output signal can be amplified by
this huge factor and as I said the Q.

in mechanical systems can be on the order
of a Towers as to a few millions so

you are resin and your.

Transfer function would be amplified by
the Q fact now how can we use this in our

advantage to cure a gyros that are not
necessarily similar to a twenty

four because in a twenty four there
are fabrication and ideologies that would

affect the frequencies of the in plane and
out of plane modes of the twenty four.

Now if you look at the degenerate
modes of a disk resonator for example.

The advantage of using a discus
structure is that there or

pairs in the pace of modes the.

That come at the exact
same Friess creek and see.

Corresponding to each other so for
example you can use most of this

type of shake shapes these are elliptical
modes and you see that the two modes

are about forty five degree apart from
each other spatially but as I said that

exact same frequency so without doing any
type of engineering by just defining a C.

metric a structure like a disk I can get
two frequencies to be equal and then there

are high ordinals you can use the third
order mode or the fourth order mode again

there especially as spatially apart from
each other by about thirty degrees or

twenty two and a half degrees but
the frequencies are exactly quote and

as I said if you the frequencies are equal
you can take advantage of the Q factor

now this is the actual mode shapes
of a hemispherical gyroscope

which is a version a treaty version
of a disk resonator now the three D.

structures This is similar to the modes
that you get out of a vine glass.

Excite it wine glass resonance by
just picking up of wine glass and

and you know hitting it on on its rim
the tone that you will hear corresponds

to this particular mode and
the problem with implementing those

A structures on a to the substrate is
that is is the difficulty of fabrication

is very difficult to get the three via
structure so to simplify things we will

use a disk resonator which is the true
to the version of those devices so

this this question either is or shown for
example here this is actually made in

silicon is it this that is in
the order of one millimeter

in diameter the disk is free
is actually suspended and

the horrors that you see here are created
to suspend a structure from its underlying

substrate but it is supported at the
middle to this pedestal so this is like.

A little.

Nail That goes down into the center of
the disk a structure to nail it down to

the substrate so it's a suspended this is
what is attaching them in the very center

and as I said you have these modes that
are exactly equal in frequency and

because of this axis symmetrical
structure the two most now can have.

Exact frequency and you can take advantage
of the very high Q. factors that come for

these modes and I'm going to show you
something you can range from ten thousand

to ten million for a structures that are
very tiny one millimeter to two millimeter

and you get these without even
needing to go to a high vacuum so

you can get it in pressure levels of or
in the order of one to ten or

you know that you know if there is
a molecule surrounding this structure

the resistance that are shown by those
air molecules can provide damping and

that damping can reduce the Q. factor so

sometimes you need to go to a very high
vacuum to be able to get high Q.'s for

this the stiffness structures you don't
need to go to a very high vacuum you can

actually get very high Q.'s by moderate
vacuum so that would give you now the high

sensitivity and low noise that you need
and then finally this the resonant

frequencies in the order of one to ten
megahertz this is different from for

example the vine class when you excite
the resonance you can hear it so

it's in the sonic range is in
the few killer Hertz range of one to

five kilohertz range these
are outside of that sonic range and

the reason we're moving up in frequency
is that we want to have a very

large bandwidth despite of
having this high Q. factor and

also have superior resistance to shock and
vibration if you had

low frequency your Gyda would also
response to linear vibration but

if you take it to high frequency it
would be a lot less sensitive and shock.

So because he has its own event and
we're going to explore that now how do

we go about making a disk a structure like
this with electrodes around that we would

use a process that was actually developed
here is known as hops process and

I'm going to give you the details of
it throughout my talk and what this

what is unique about this process is
that we use the silicon on Sci Fi for

the disk is actually implemented in the
device layer of that is so I have a for

and now we need to have very tiny
electrodes tiny gap Aleck throws

that you know generate a large
force into application

of a voltage to the electrode on the body
of the disk so when you apply and

like it will tissue these electrodes it
has to apply a large force to excite that

resonance of high frequency and
as such we need to have nano gaps so

that's basically the secret sauce of this
process how can we get gas either in

the order of a few hundred
nanometer between.

Silicon structures that are electrically
isolated from each other.

So.

In actuality the motes that we use are not
the first order elliptical modes and

the call to two that I showed earlier that
actually any call to treat the general

modes so there's a higher order most and
these are animations of pain from manses

highly exaggerated of course to disk
would never be form as much the vibration

would be a fraction of that gap size and
I mention to you as to the vibration in

practicality it would be in the order
of few tens of nanometer at most.

And what's going to happen is that we
have electronics that would drive this

disk resonator into in into
this particular resonance and

control its vibration amplitude
Now once we apply rotation

to this device what's going
to happen is that the.

Coriolis force causes energy from
this mode to its sister mode

which as I said is that exact same
frequencies just a spatially or

toggle to it is that about thirty degrees
and what's going to happen is that now we

would detect the motion of the Second
World by placing the electrodes for

the N.T. nodes of this mode are and by
processing these signals to low power and

low noise electronics we can actually get
a measure of rotation for this device so

overall is not that hard to figure
out how it is operating but

to get a pristine signal out of this
device is extremely difficult so

one of the difficulty comes from
the fact that I will be used to disk and

I in an ideal world of the two most
would come to be exactly will.

Disappear Yea sions the two morons
are not going to be exactly equal so

they're going to be have a difference
in magnitude and in value.

Dat you know the Dell toll make us
of the tool makers are not equal and

that can be caused because of and
I saw that this if he

did last this is the of the two most
can be a little different or and I so

inertia the mass the effective mass of
each one of us can be is just a tad

bit different because of process fairy
ins and we talked a few years so

ask fabricated this is what you're
going to measure out of the device so

that the red line shows the first mode and
the blue line shows

the second mode of resonance and this is
taken out of the network analyzer and

you see that instead of both being at
seven point two five megahertz or slightly

apart so how do we fix this because we
need the two to be exactly overlapping

river we use the electrodes that
are around the structure to apply D.C.

voltages to to turn to modes and this
is enabled by what is known as an extra

static spring softening so
by applying a voltage V. apply a.

For the worse due to the none
the near three of the electricity.

That acts like a negative a spring and
that negative a spring now can actually

loosen up the stiffness of this higher
frequency mode so it did it it is now

brought in on top of the first mold so
the frequencies need to be matched and

the other thing is that in this each
one of these plots you'll see is

small reminiscence of the other mode on
the first note and this is because the two

modes are not or toggle completely if they
are completely orthogonal then you excited

one of them you should not see a copy
of the second one at all in that

spectrum the reason you are seeing it
is because the more that as I said has

some non ideologies so
you not only have to.

Make the frequencies equal you have to
make them orient the same they do in

the right direction you have to
make them more time going on and

that is done using other set of electrodes
around the structure so all together

you have to do the coupling of the modes
as well as matching of the frequencies and

by doing that you can get the two
most to be completely matched.

Now it comes factor so with this cue
that you have been talking about Q.

is a measure of energy
dissipation in a resonator.

And it is a combination of different
effects it can come from squeeze

film damping which is as I said the effect
of air molecules surrounding the structure

providing the resistance to
the resonance it can be due to key or

terminal lastic damping this is the
interaction of elastic fields with Turmel

fields as the device is the firm
it gets hotter in some areas and

cooler in other areas and because of that
there is a heat flow inside the structure

the interaction of these heat flow with
the mechanical resin can can produce

some additional losses anchor losses
this is a measure of the loss of a.

Energy to the mechanical anchors So
basically the acoustic energy that gets.

This is painted by going
into the substrate.

So that can have any a huge impact
on the Q Of the resonance and

then you have surface losses.

As a lot of surfaces and the surfaces
are not ideal a lot of roughnesses or

dangling bonds you can have some energy
dissipation through those surfaces and

then finally intrinsically is our losses
that are inherent to the material of

the resonator you know some materials have
higher losses compared to others Diamant

for example is a very low loss material
silicon is also lost but not as

good as diamond for example so if you go
to Diamond ultimately is possible to get.

Like there's so all of these they need to
be suppressed to get to the very high I.Q.

factors that you would like to get and
remember the sensitivities proportional to

Q So
talking about the squeeze film damping and

then I think is that each one of these can
more or less be modeled these days using

the multi physics simulation platform
when it comes to squeeze from damping

you'll see that it has a strong
dependency on the turret power so

the gap so if the gaps are is small and
we talked about having nano gaps

you've will have a large damping
squeeze film damping in these devices.

If you want to have less damping you're
better off going to two larger gaps of

this but we need this small gaps to
be able to actually use this mode so

that the way to to actually
eliminate this is by

reducing new effective and.

Which is the effective vehicle
lasting damping coefficient

that has to go to higher frequency
modes so that's something that we will.

Going to be taking advantage of it will
go to a higher frequency to reduce.

This damping mechanism and also

go down a little bit in pressure
to reduce the new effective.

Temper lastic damping.

Is a complex mechanism that depends on
the core fission of terminal expansion and

the turmoil modes inside
the structure again nice thing

about it is that you can model it very
easily and you can engineer to be

a small value for these particular
modes the modes of electrical

modes of the disk resonate there are ten
lasting damping is typically a smaller

compared to when you have Flexeril
modes for example in twenty four crew.

It's another advantage of using stiff.

Modes are they of a disc resonator they
they have much to small their T.D.

comfort the tuning forks and then finally
anchor last this is very important

you know we use a video a small
pin to support the structure

this is you know suspended this and
attach it to the to the substrate and

there are big dependencies on the actual
physical size of that supports

when it comes to the anchor last so
really have to engineer this and

use the right modes again to get
the highest Q. anchored as possible.

And this you know the nice thing about the
any call to tree mode is that as you can

see in the middle you won't see
much interaction the stress fields

in the middle are in the blue collar which
means that the stress is very minute.

That is read so putting an anchor in
the middle for any call to tree mode

which translates small anchor losses and
that's the reason behind why we use and

cool to trim So all together now
I love the shot shows you that

this resonated provides only
a axis of rotation around its.

The normal to the plane of
substrates direction so

if you crow to the round around that
normal you'll get a measure of rotation

if you're interested in trying axial
rotation measurement so what we do here is

that one way of implementing a try Axel
gyro is to flip the die and mounted.

Inside that package now if you have
a small Di This is possible and

these guys are pretty small investor level
package form we're talking about one

point five millimeter by one point
five millimeter a very tiny Di and

they take a sub it is
about one millimeter So

it is possible to actually flip
it mounted on its side and

then use what is known as Corner
bad bonding which are actually

the Via bonded by bronze
electrical connections

from these pads on to the substrate and
then further on on to this AC So

now you see that this is this seems to
be very complex this system can be can

fit into a almost five millimeter by five
millimeter package and can be produced.

Are compatible with consumer electronics
where you get going into a cell

phone Now obviously one of new way
from a situation like this and

actually come to a single
chip solution for

demands I'm going to talk about
that later today now in package for

these devices show a queue of
seventy seven thousand again.

Rather is that you typically get out
of a C. mosque for a high I.Q. inductor

you're talking about a Q. of one hundred
is the maximum we can get out of the C.

minus so mems doesn't offer orders
of magnitude higher Q. factors or

lowered loss compare what you are to
what you can get out of a process.

Now let's take a quick look
at performance in the past.

I rose so one very important measure of
performance is what is known as motional

impedance so this is the impedance of the
resonator at its resonance frequency and

what you see here is that it has a complex
dependency on the resonant frequency Q.

factor and more importantly the fourth
power of the gap size that's why we need

to go to very small gap sizes because we
need to have a small motional resistance

so ultra small cap apps are needed for
this particular

purpose a scare factor R.V.
losing anything by going to high frequency

The nice thing about the scale factor is
that it is independent of frequency so

if you go to a higher frequency you should
in assume that you're going to get a loss

of sensitivity because this device
is pretty stiff it doesn't move much

while it is working based on the coupling
of the two most not exactly how much

you're driving it so
this sensitivity is proportional to the Q.

so the higher the Q.
the higher sensitivity and

inversely proportional to the gap
size mechanical noise this is

the ultimate noise mechanism the terminal
noise of the mechanical element

itself that would determine the minimum
resolution of the system so

that one is inversely proportional to
the Q factor resonant frequency and

mass so we are going to gain through
higher resonance frequency and higher Q.

are beneficial and
then finally the bandwidth of this

this is the treat the bandwidth of
resinous the benefit of the sensor and

they have band with a resonance is given
by the resonance frequency divided by Q So

going to higher resonance
frequency would enable us to get

a larger bandwidth despite having a high
I.Q. so are together you're not going

to lose anything going to high frequency

you're going to actually gain
a lot by doing that now.

Previous to slide I showed you the Q.

of those of A for level package parts
to be about seventy seven K. so.

What is the limits of Q Can we get to
much higher Q. than seventy thousand and

the end yet the answer is yes the one
that you saw it was limited by ten lastic

damping in a perforated disk
resonator Now those perforations

those hurdles are needed because of
the release process we had to remove.

That was underneath the disk
resonator to suspend

if we come up with a different process
flow such that we can use a solid disk so

this is a solid this kind of two
millimeter in diameter that is actually

design a fabricated by my poor solder
Amin who's sitting in the back and and

they're using a solid disk is that you
don't have to have those release holes and

those release holes typically induce
a lot of time when asked again pain

now by eliminating them you can read it
get rid of that him with acid damping and

get to Cuba to use that
in excess of a million so

this is measured in our
lab this is the prices for

fabricated in the clean room here at two
point seven megahertz this shows a queue

of about one point three
million ride the two modes or.

A a you know a smaller
split that can be tuned by

applying electricity forces now
we're very excited about this

he's going to talk about this at
the transducers conference in June and

I'm just showing this to to highlight
the the advantage of this technology

the very exciting thing about it is that
the now actually get to measure a Q.

that his visit to the simulated value and
simulation people

are predicting to get a T.D. plus a Keizer
and he's there is a measure of the at

internal intrinsic type of loss
that I mentioned to you airily or.

That that in simulation predicts a value
of about one point five million for

Q and look at what you're
measuring very close to.

Eight million so this is I think about it.

And now a little bit going back to
the process flow hops process by the way

this harks distance for highest peak
ratio poly a single crystalline silicon

technology which basically enables
fabrication of three D. silicon.

In a in a process technology while having
a scalable air gap between the electrodes

so we can get nanometer scale air gaps
between the ten to hundreds of Micron take

a structure so this is the poly electrode
that I was talking to you for excited and

sensing of the disk resonances this is the
body of the disk forty Micron take S.-Y.

and the gap size is about one
hundred eighty nine a meter.

Now this is truly the next
generation members process which

eyes reduction because enables us
to go to the higher frequency and

the smaller resonators with
stiff block acoustic modes and

be able to excite them is
actually them and sense and so

it has this scale ability that is often
associated with seem OS seem OS has

basically enjoyed this longevity
because of being a scalable and

here in the memes domain to a level
technology you have to have a process

where the capacity to services to
the scalable so hence having that

small gap size is important now how are we
getting this the small gap size most of

you probably are familiar with the default
I.E. process deploy process you

is based on sequence schol etching and
passivation right so

is limited by first of all the resolution
of the toddler feed because you have to

create the mask first so you have if you
want to go to an animator you have to use

nationally telegraphy which is expensive
and also it is limited in his aspect

ratio the best that you can ever possibly
get is seventy to eighty to one butt.

This is not with the development the tools
we have here is the tools we have

here is about twenty or thirty to one
you get the biggest problem here is

that these are scalloping is are going
to cost in reducing the gap size

now in the hops process let me do is
that the first two hydrogen annealing so

that hydrogen annealing would
actually cause flow and

to smooth the cycle so
we get rid of those a scallops and

then we do a sacrificial oxide
formation on the side walls and

this is typically done through Turmel
oxidation term oxidation is the same

process that defines big
big the gates of a mosque

transistor with great accuracy you know
that the techniques of the gate are.

Actually now at a few nanometer in
the most advanced CMOS processes so

we have a lot of room to to do to skill
the gap down we can go to a few anatomy

there if you're far away from it if you
have to transfer it's probably silicon so

after that to be removed this
oxide in a release process

you know we can use vapor H.F.
or just where they are and

now you have a you have gotten
yourself a gap a vertical gap

that is not limited by Aspect Ratio
of the deep are a process.

Really nice you don't have to use natural
it's hard to feel you can use software you

know Mike on a target few sizes and then
the other thing is that you can also have

electrodes that sense the out of plane
motion of the resonance by putting

electrodes on top of field
trenches that are filled with

oxide in this particular case and
the oxide is lay there on the move so

you can get gaps in the vertical
direction as well as lateral direction.

Now what is the best thing
that you have achieved here

Robbie use this particular device
which is called the silicon Valcke

acoustic resonator is a is
a rectangular bad or that vi.

Actually like to excite it's most
is the stiffest mode which is the.

Expansion and

contraction expansion and contraction
these are very very stiff modes.

So the best gaps if you have so
far been able to get reliably between

you know finished surfaces
about fifty nanometer.

And this is done on a twenty Micron take.

So we can layer so the aspect ratio of.

Growing fifteen enemy there to twenty
Micron is you know almost approaching

thousand to one thousand is going from
fifteen enemy that to fifty microns so

it's a little less than thousand and
one and what is enabling this or

what this feature is and
they will Ling is that now this device has

a resonant frequency of about one hundred
fifty megahertz we can get a queue of

fifty thousand by applying a very
small voltage to it this is the D.C.

voltage that is applied to the body of
resonator two polarizers transducer D.C.

voltages are actually used to excite this
resonance are in the order of humility

Wilts and emotional impedance you
probably can't read is a measure of

the insertion loss of the device
essential loss is about five D.B.

with which is very respectable is similar
to what you can get out of course devices.

If you were to use this.

You know to make a resonator the DAB the
hops process just by dry a change you have

done that also this is another
side bar the etched dry

at two hundred nanometers or gap but we
can only go down to about two microns so

the gap aspect ratio is ten to one here
compared to about two thousand to one in

the previous life and and
you had to use National a TA defeat

to define these dimensions and
read the measured was.

Your Hertz we still get very hight.

Hughes but the big problem
is this is social loss which

as is shown here is
minus seventy sixty B..

Fifty even worse than what you saw in
the previous a slide which corresponds to

five years of magnitude larger impedances
So that clearly shows the advantage

of having nano gaps in memes resonators.

Finally visceral level packaging so
we've finished the hops process here

how do you create that
moderate vacuum that I talk.

So what we do is that we use

is to get electrical connection into
a Silicon Valley for so we edge

trenches into a low resistivity silicon
and then fill up the trench for

the oxide we bonded through the main
of a for and then polish it down so

that we can access these these vias and
then be put down rock layers on top so

oftentimes the processing of
these fears he kept me for

is actually involves the same
number of steps if not move or.

For itself so it is a pretty hefty
section of the process flow itself and

then finally this is the V.A.
for that is be that is ready for

deicing in single Eish and
be interface with the electronics.

So one thing that would be beneficial
is to get rid of this capping wafer

because as I said it is often the same
cost as the base pay for itself so

whether if we kept the memes of A for
the sea Moss way for if it's going to

interface with seam OS Why not just kept
it would seem awfully for and that's.

An Every do that if possible
the problem is that

there to be a first usually the die
size is don't match each other so

the demands that is typically a smaller
compared to the A six size and

you end up you know trying out
a whole bunch of a whole big.

Of the many safer because
of the size mismatch so

how do you answer how do you
address this problem the add to.

To go to integration so if you have
more and more components on your men's

you know if you need them but
the different axes of acceleration and

rotation together on a bigger members die
then you can come to a level that you

almost match the memes that die size
of at the scene last I size and

that way you don't have to throw out
a big portion of the members of a for but

overall is a great view of capping
the memes because first of all you get

rid of a lot of these feet through and
pester the capacitances that

are present if you have to
use it too cheap solution.

And getting rid of the feet through and
coupling although you can do it with

sophisticated I can the electronics and
do a lot of you know digital

correction it usually translates to
a lot of power consumption so for

internet of things when people are taking
a bath using sensors with a little power

you have to reduce power you have to
get rid of these pesky capacitances so

it becomes particularly important
when power is of great concern and

at the same time not only just reducing
power it improves the performance and

the precision of the system because now
you can better match the face delays

that are very important when working
with high I.Q. resonances and

the modulating the signals So overall it
would enable single chip self calibrating

gyros that can deliver the performance
that are needed for navigation.

Now in terms of performance
where are we today

this is the measured performance of a bar
gyroscope with the angle random walk

that is limited by the turmoil noise of
electronics to about point thirty sixty

proved our gyro itself
the mechanical element is still far.

Our firm was the measure because
of the electronic noise so

you see the impact of electronic
noise here and the bias is to bill

it is about three point five per hour
really need to be to enable navigation for

indoor applications we need to be
had about one degree per hour and

we are caught in the I think there.

Reason is that this particular device
was very small in size to enable the die

on edge assembly if you go into a fully
integrated single Di conflagrations with

all axes integrated on the same die if you
don't have to use such a small device be

can go to a larger device and
improve the performance but in terms of Q.

factor over temperature so this is another
important thing that the device has to

work well over the average temperature
ranges and all the actual conditions of

use them including vibration The nice
thing over temperature is that the Q.

is very predictable it has a very
predictable behavior the two modes have

equal Q. factors and they are following
each other as the temperature changes

which is excellent that means that we
don't have to do much compensation.

And this is now due to the fact that Q.
is designed to be ten more lasting

damping limited until last examine
has a temperature behavior.

Now because of the sour the state nature
it has a great immunity to shock and

vibration you have compared our
device to a number of consumer

TS G.'s that are commercially available
as well as an industrial T.F.T.

and what you can see is that all
response to random white ration.

Which you shoot your perfect rejection to
it is a lot smaller than the other parts

of the are available in the market and
in responding to shock this is forty G.

of shock the shock will
induce an offset shift if you

produce you know if you have exposed.

Last shot it would cause a shift in
the offset and the I told you about

the effect of that shift it will add up
to the errors so the bad is really raw

it's that's have a very small shift
in offset with respect to the.

Shock as well these numbers are not
good enough to enable navigation so

as together.

You know what's going to happen is that
many you integrate that outfits the output

of the gyro over time if you have
a large offset you're going to see that

the out with will grow over time if you
doubt even any rotation being present So

this is without rotation you see that
the Off with his growing over time.

Showing dad that is rotation while there
is nothing actually out there there is no

rotation and you know this is
somebody is also there are actually.

Read the collected out of our devices that
if you use it for indoor navigation then

you use a good vibration immunity that is
offered by our device used you see very

much better results compared to devices
that are that have poor vibration immunity

and this is just when you have rocking So
the walking has an effect on the gyro

I have a video here that is
produced by our one of our.

City partners.

That is planning to put this gyro into.

Motorcycles and you know in motorcycles
there's a lot of vibration of course

in any so this is the response of
our gyro and a competitor gyro and

you see that even that isn't the libration
the gyros work very well and

now he's going to turn on
a vibration stage under the lice and

they see that the device Bishan would
cause this device this well the cycle now

go haywire and that's the actual
operating condition and that.

Shows you how important is to have
gyros at their vibration resistance and

that's why we don't have any of those
drivers available for navigation.

So.

We can do better.

Than Once you have the gyro
working by electronics so

this is what I meant by brains of
the Microsystems the CMOS can come and

do a lot more calibration including
self calibrating the gyro sort of Eakin

actually apply signals that would
mimic the effect of rotation and

measure the scale factor of the device in
situ so by doing that if there is a shift

in this scale factor over the lifetime of
the device we can measure it in situ and

correct for it and by so that would
maintain the performance of the device or

its lifetime more important than
the scare factor calibration is

that you can also calibrate
the bias out of the device and

this shows you the Allen deviation
plot for a structure that doesn't have

any bias calibration you see that after
some point something off the sun time by

one to ten second the device would
reach its minimum noise level but

then start screwing up it
becomes less accurate because of

you know temperature and
vibration of facts if you use this self

calibration take me if you can get rid
of these effects of temperature and

why ration and produce a bias for that
is constant stays constant over time and

that time can be extended two hundred
seconds and even hired two thousand second

which correspond to a few minutes of
navigation This is the work done by my.

P.C. issue than they're asked was also
sitting in the back of the room very

impressive now talking about going single
chip though how if you don't want to

mount the Dan it's edge how do we get X.
and Y. gyros on it plane or coffee ration.

The answer is that we use other modes of
resonance of eight annulus a structure so

in this particular device I'm sure I don't
know why it's showing the other modes.

So this was supposed to be the annulus
Well maybe I don't have the right video

files in here but it's the same type
of mode that is excited in playing for

this annulus mode I'm sorry
I don't have the modes so

I maybe I just don't play these so
that you can see the modes from here.

So the other for the inflamed
rotation the X. arrived with patient

would transfer the energy out of this in
plane mode into an out of plane mode.

Then annulus So if you have electrodes
places strategically on top of then

you list you should be able to pick
up that resonance and measure the Y.

rotation so that's the technology that we
will use the device that we will use to go

to a fully integrate that
I am you want to chip.

The device we have made it here.

It was a V. for a little package at
the foundry that works with my start up

company and this is the response
of the device to an X.

and extrication we can
actually rotate the device to

fit in ninety degree consecration to to
make a device sensor and opposed to an X.

sensor or we can combine the X. and Y.
together into a single a structure but

it has some cross coupling issues
of you're looking at the moment and

it takes us a little longer
to get it to market.

But all together using Hartson R.V.
can do a gyro the X.

gyro RA and then the three axis
accelerometer all integrated onto

a a single di di sizes for
a millimeter by six millimeter

of course is bigger than the one that I
showed you earlier that was only one and

a half in the way that we want to have
millimeter but it is a rather blank.

This space here in the between that
we can optimize the reason for

this blanket space is that
the ceiling green is of the V. for

level casting process comes around
each one of these devices to.

Individually package them we don't need
that we just didn't want to change

the process flow at this point in time to
accommodate a larger cavity size because

every time you change the cavity size
the foundry has to really opt on being

pressure to get a good yield so we just
piggy back on the existing process and

that's why it is the die is larger vs done
some work to show that this stuff I can

come down to a two millimeter by two
millimeter in a free quadrant consecration

where one millimeter by one millimeter
would be for each one of these devices and

that the three axis accelerometer
would be the shrunk in size and

place in a one millimeter wave
one millimeter area the Q.

factors are still very high you see the Z.
has the highest Q. but the X.

and Y. gyro have also very high Q.'s for
the plane modes in the.

Ten to twenty thousand you also have
a timing resonator integrated here

that shows functionality and high Q.
resonance this is too integrated that

not only I am you with also the timing
element with the diet you so

a little bit about the accelerometers
these are single proof mass single axis

accelerometers So for three axis you need
three proof masses one in the X. one Y.

and one in the vertical direction
these are the moments that use for

sensing acceleration accelerometers
are more mature devices the real challenge

here is that we need to get them to work
in reduced pressure environment because we

do we do go down in pressure to about one
to ten Tor to increase to improve our Q.

factors and actually a lot of those being
a static devices they don't like I.Q.

because they tend to.

You know overshoot and.

Sort of have downs that are detrimental to
the overall response time of the actual or

ometer So turns out that if you
reduce the gap size you saw in

the squeeze film damping a slide you will
get more damping So we're going to take

advantage of the small gaps here in
Axelrod meters to increase the damping and

reduce that during down time to get that
weather response time out of the device so

this is the electrode here you see
in production you see a reminiscence

of the probably on the poly This is
a pretty thick poly six micron and the gap

size in production very pristine gap sizes
in about thirty two hundred nanometer

the response of the accelerometer again
the pretty tiny one two by two millimeter.

The response of a.

This is an X. axis response was not
showing in the response for the RA and

Z. axis excellent cross axis rejection
which is again important for

navigation applications you don't want
to have a lot of cross axes sensitivity.

And this is the other plane
accelerometer again showing

good out of plane sensitivity and
no sensitivity.

Liberation.

All right so I'm going to know I'm
going over time so to conclude.

I hope you guys can see that short range
strap down navigation is a killer opt for

high performance gyroscopes and
in general.

The application is is very important and
is also.

A grave impact imagine having to take
the same type of navigation to enjoy it

through G.P.S. endorse firefighters can
use of immediately if for anything else

and then eventually we all can benefit
from it a size of it is humongous G.P.S.

is every verse this would be parallel
to G.P.S. whatever you use U.P.S.

you will have these things and
video body aches coming into.

Fruition and the ending up using your
bots anywhere they will have to have or

homages into the need to use.

Homogeneous tri axle gyros whenever
form of lot most importance

homogeneous meaning these are devices
that are actually symmetric Z. gyro for

example is axis in metric the average
plane ones are not symmetric because you

use an in plane mode and then the plane
mode and the out of plane will is affected

by the technical variation of the S.
Y. so small by size is key high I.Q.

is key and
large dynamic range is very important.

And so is why ration immunity and
that is all offered by barge I rose for

level packaging with TS fees have
been enabling to get to market but

we need a smaller pest and
it all goes to small passages and

they were by the better tiers you
processes or we would just go directly to

capping with a secret.

To get fully integrated ten degree of
freedom positioning and navigation micro

system positioning navigation and
tying the micro system.

And I show you showed you what those ten
degrees of freedom on are temperature and

process composition are very important and
hopefully those will be done.

Now liberation techniques that
will deliver performance.

So I'd like to acknowledge
the contributions from my past and

current students and postdocs without them
this would have been impossible obviously

enjoyed working with great students and
post-docs and then support from DARPA and

N.S.F. throughout the years has
enabled this this worked thanks so

much and
I'll be happy to answer any questions.

I mean.

I mean after the first
one before I forget so

this is done by just nonbeing that
capping of A for the base pay for

in in a one two or pressure level so

that's the pressure level that is
provided inside that bonding equipment.

So there's no get there or
anything special.

Done to maintain that.

Yeah.

Yeah.

Instead of crystalline silicon.

Paul Levy mostly use it for electrodes.

So it's not part of
the the resonator structure.

It would like to use single crystalline
silicon for making the resonating

a structure because material properties
of single crystalline silicon are very

well known first of all and
are very repeatable.

We don't have to worry about the quality
of the silicon deposition so

if you don't want to use poly as a
structural material for the device itself

just for
the electrodes there is no drawback.

And that's what we do but

even if you decide to use poly you have
shown as a structural material for

the resonator you have shown
that poly can be very high Q..

As high of a Q A single
crystalline silicon

The only thing is that when
we distill feeling processed.

We use we may end up having a void
between the two sides of the party

a structure as the joint and then I
race against relive that really it's

not important to us because we use it as
an electrode and we don't care about it.

What again we have shown that we can get
rid of that group if you end up using get

at some point in time for
the structural material of a resonator.

So lever all.

I think this is a again and again another
plus for this technology that not only you

have access to single crystalline silicon
you have access to probably crystalline

silicon for the structural or
non a structural type of elements.

And run big advantage of poly crystalline
compared to the single crystalline silicon

is that it is isotropic single
crystal silicon has it's an isotropic

which when you're trying to do acoustic
design makes things very difficult.

If you use an isotropic material like
Poly things become a lot easier.

So it has if you have access
today to a lesser I.V.

for the device layer of poly silicon
instead of silicon we did reliable and

repeatable material properties we would
probably use that is just not available or

if they are.

They are very expensive and.

They may not be reproduced reproducible
is as good as what's in the crystalline

is.

You know.

Hydrogen and using right.

Process.

Yeah so that hydrogen anything.

It's actually a.

Is an optional the step that you can do or
you may not.

Take it man it up it is
basically done in a hydrogen

furnace at elevated temperatures
about Italian degrees.

Again you can play with the temperature
thousand to eleven hundred sometimes

even a little less than thousand and
is done for a very short period of time.

Between five to ten minutes and
what's going to happen is that

silicon with a start to reflow and
we have shown that you can get rid

of the scalloping as well as
the strike is on the side of all.

With it if you if you do it for

too long you will end up sure
reshaping the silicon and even for

various small short period of time
when you have an interface of silicon

the with another material like oxide
you will see some defamations there so.

If for processes it does
require some tweaking and and

you know fine tuning to
be able to use it but.

Use for getting rid of any type of
irregularities on the side wall

which mates end up being
a liability in terms of.

Long Your mechanisms.

It's just better to work
with mirror finish surfaces.

Thank you very much.