Implantable Networks of Wireless Nanoelectronic Devices
Sep 5, 2023
For more information, visit https://nanohub.org/resources/20234.
My name is Pedro Eredope. Professor Eredope is a current associate professor in the School of
Biomedical Engineering and Computer Engineering at Purdue University. He is also the director of
the Center for Incredible Devices. He received both his Bachelor of Science and Master in
Electrical Engineering at the University of New Hampshire in 1997 and 1999 respectively,
and he received his Ph.D. in Neural Engineering at UCLA in 2003. Currently, Dr. Eredope's
research is pursuing a set of tools to find a modular approach to the design and biological
input for clinical treatment of psychological disorders. Today, he's going to give us a lecture
about implantable networks. Please join me in welcoming him.
Thank you very much for the kind introduction. It's very nice to be here and talk about some of the work
that we're doing at Purdue in the Center for Implantable Devices and more broadly in the department.
I'll begin by going through some of the work that we're working in particularly in my group, but then
I'll touch on some of the things that we're interested in more as a department and in directions for growth
in the future. I know most of you are graduate students as you're thinking about post-docs and as you're
thinking about faculty positions and so on and so forth. We have 170 faculty positions open in
Neural Engineering right now, so there's opportunities. I'll talk about that towards the end.
The motivator for this is that the medical marketplace is about 19.5 or will be soon about 19.5% of GDP.
That amounts to about $3 trillion a year annually. That's just in the United States. Yet the way that people
treat medical conditions has remained fundamentally unchanged the last 100 years in the sense that
every individual doctor works on an individual pathology. So your cardiologist works on your heart,
your neurologist works on your brain, your gastroenterologist works on your gut, and those three doctors
will never talk to each other and they don't particularly care what's going on in your heart or in your gut
if they're working on your brain. These different body areas or organ areas are treated independently
from one another and the methodologies are often at odds with one another. So our sense is that we could enable
multi-pathology treatment, so treatment of multiple different locations in the body through a physically
distributed network of nanoscale wireless sensors and actuators. So having a distributed network
throughout your body and your brain, your heart, your gut, wherever, that monitors and stimulates,
that senses and actuates what's going on in your body would deliver fundamentally a different way of thinking
about medical treatment and eventually a paradigm shift in how medical treatment is administered
and performed. And so in order to do that, we have to develop a number of different technologies.
We have to develop new ways of sensing, new ways of actuating, new ways of networking devices located
in disparate locations, and new ways of miniaturizing these devices so that they can go wherever you want
them to go in the biology. So this is a collage of all the people that work or worked at one point.
It's a little bit of a data collage now in the center and some of the devices that we've made.
But this is an enormous group effort. It's not just one person. It's all of these efforts are bound to be.
So I'll begin by going through some of the sensing modalities that we do, and this is really just an overview
of what I could fit into a 45 to 50 minute talk. But I'll talk about some of each of these different categories,
sensing, actuating, networking, and miniaturization, and try to give you an overall picture of what we're trying to do.
So on the sensing side, one of the projects that we're involved with is glaucoma.
And in glaucoma, what you want to do is you want to be able to measure the pressure inside of the eye.
But the best model for doing this work in humans is actually not the rabbit, which is what most people think
when they think eye surgery. They think rabbits as an animal model. It turns out the rabbit eye is quite different
from the human eye. The only way in which it's similar is the size. It's about the same size,
and so it's nice to work in from that perspective. But in terms of the anatomical structure,
from an evolutionary perspective, we're much further from the rabbit and much closer to the mouse,
which is a pain in the neck because mice are really small. So we had to go and come up with a way
of measuring pressure inside of the eye of a mouse, which required us to make an entirely different, new sensor.
And so this is the sensor that we came up with. This was a collaboration between Professor Bill Chappell at Purdue.
He was leading on this sensor development and myself through a Howard Hughes-funded effort
in partnership with the Jackson Laboratory and Simon John out there.
So this sensor is a capacitor pressure sensor. It's 500 by 500 microns.
They can become as small as 300 microns, as big as 800 microns, but it gives us the sensitivity
of about half a millimeter of mercury that we would need when coupled with an integrated circuit
to measure and log pressure changes in the eye continuously over time.
So it allows us to track the results of any kind of therapy, which we might want to administer for glaucoma,
like drops, drug therapy, or surgical intervention like a shunt that you might put in,
or a laser trabeculectomy where you modify the biology itself to increase the amount of flow out of the eye.
Another area in which we wanted to get involved with sensing, and this is actually our bread and butter,
is in the neural domain. We wanted to be able to record individual unit activity
in various different locations of the neural cortex.
And in order to do that, you need these implanted depth electrodes,
and here's a collage of various different commercially available electrodes made.
University of Utah, which is now a company called BlackRock, the electrode making portion of it is.
The University of Michigan made these electrodes, these and these, that are now manufactured,
and these that are now manufactured by Neuronexus as a private entity as well.
There's a few other designs out there from UCLA and Drexel and a couple other places.
All of these electrodes have the same problem, which is that they don't work.
There isn't a single commercially available neural recording electrode that works.
There never has been, and that's interesting because you don't often get to make absolute blanket statements of fact in this world, right?
This is one example in which you can. You can actually say all neural electrodes fail,
and the reason they fail is because when you put them in the brain over a period of time,
you can see that a scar forms around the area where the electrode is,
and that glial scar changes the impedance and the signal-to-noise ratio of the interface between the contact point on the electrode
and the cell that you're trying to record from, and it happens all of the time, right?
So electrodes fail. Every now and then you see a paper that has electrodes that have been in a monkey,
and they've been in a monkey for a year, and they still work.
The reason that happens is because you put 100 electrodes in, and you do this in 10 animals,
and you'll get one animal with two electrodes that work in a year, right?
So the yields are so catastrophically low that they're good enough to get a neat paper,
but they're useless in terms of developing a device that's clinically relevant.
So how do you make something clinically relevant?
Well, my answer was to just avoid the problem altogether.
I just thought, you know what? Heck with that. I'm not going to do electrodes
because Michigan and Utah have spent tens of millions, literally tens of millions,
of NIH and DARPA money over the last 30 years making these electrodes,
and they can't make anything that works because they're all, they have to be a certain width
in order to have the rigidity to hit a deep brain structure,
and once you have that width, you cause that injury, and that's it. The game is over.
So I got to Purdue, and I said, I'm going to make everything that interfaces with other people's electrodes,
but I'm going to make the electrode problem somebody else's problem and just steer clear, right?
I'm an assistant professor. You don't want to tackle insurmountable problems.
So then this guy, David Yarrow, who wasn't even in my lab, wasn't even in engineering.
He's over in ag school. He walks into my lab, and one day he says,
Professor, actually walked into my office, he says, Professor, I've solved the electrode problem.
He wasn't my student. I didn't even know he was.
And my first reaction was to say, get out. I'm busy.
But then, you know, you think part of your job as a faculty member is to mentor
and nurture young minds and all that good stuff.
And so I thought, all right, I'm going to listen to him for five minutes,
and then I'm going to tell him to get out.
And so I said, yes, David, all right, whatever you come up with.
And he said, well, the problem with these electrodes is that they're really big.
So all you've got to do is make them smaller.
I thought, yeah, you're not the first to think of that one, right?
So he said, no, no, but the literature shows that if something is smaller than about 20 microns in diameter,
20, 25 microns in diameter, that the brain will actually ignore it.
It won't treat it as a foreign body.
And so all you've got to do is figure out a way to put that deep in the brain,
and you're done.
I said, yeah, but that all-you-have-to-do part, that's the crux, right?
Because anything, when it's 20 microns wide, I don't care how rigid or strong a material, is flexible.
Even glass is just flexible at that thickness.
So he said, well, yes, but what if instead of pushing it in the brain,
we were to coat the tip with magnetic nanoparticles,
and then we could use a magnetic field to pull it down into the brain?
And I thought, well, that's kind of neat, right?
So I said, all right, well, let's try this crazy idea of yours.
And so this is in my lab.
This is after a few months of development.
We made a jelly mold.
We got a 20,000-frame-per-second camera.
We made a rail gun.
And then we tried to hit a one-cubic-millimeter target five centimeters down into the brain.
There it is.
And shot at 20,000 frames per second.
There's a 20-micron electrode going exactly where we want it to go.
Now, in the interest of full disclosure,
I'll tell you that we shot this 12 times before we got it to go where we wanted it to go,
because the aim was terrible.
But this was enough to then go to DARPA and say,
I need half a million dollars to make this a reality,
and they gave me the money because they're nice.
They're not, but they're very helpful.
So they're nice in a way.
And lo and behold, here we had the electrode.
So it was a fantastic, fantastic opportunity to develop the technology.
We implanted it in animals.
We've had them implanted in animals for 90 days at a time.
And unusually, the electrodes work 100 percent of the time.
The catch is that they're still not very accurate,
and so we're working on making them more accurate.
But that's another example of a sensing technology.
On the actuating side, we spend a lot more time on actuation than we do on sensing.
And on sensing, we tend to partner up with folks in the bio-mems world to do things.
But on the actuating side, the problem that we observed was a different one.
So we worked with a company called Cyberonics,
and they funded us heavily over the last five years
and generously.
And so the work that we've been doing with Cyberonics
has been focused on trying to understand how their device works
so that we can then make it better.
So what they have as a device is they have what's called a vagal nerve stimulator,
which is a nerve cuff that wraps around the vagus nerve in your neck
and electrically stimulates it.
And that stimulus somehow goes up to the brain and does something
which makes seizures happen less often.
And I'm not dumbing it down from you.
That's as dumb as I've got it, right?
So when I first started working with them, I said,
so how does your device work?
No idea.
So I said, all right, okay, so you're stimulating the brain.
So how do you set the amplitude, right?
How do you set how much current you're pushing into the body?
They said, oh, well, that's easy.
We have the patient.
Once we've implanted them with the device, they come back in to the neurologist's office,
and then we turn up the volume on the device until the patient starts to twitch,
physically, visibly twitch.
Then we back it off a little bit, and we leave it there.
I said, okay, so what amplitude is that?
No idea.
Is there any change from patient to patient?
Probably.
I thought, okay.
So, all right, so that's a little scary.
And I found out later that Medtronic, with their deep brain stimulator,
they do exactly the same thing.
Everybody does exactly the same thing.
So I said, all right, but the waveform, I mean, surely the shape of the wave,
the duty cycle, the frequency, these are things that have been studied.
What are they and why?
And the first answer was, well, actually, we don't know what they are.
We'll have to go look that up.
And it took a while before we were able to dig up what the waveform actually was,
which turns out not to be what it was supposed to be.
However, it's been approved by the FDA.
And so I asked them, I said, well, why this waveform?
Why not a different waveform?
And the answer was, this waveform works.
If we raise the question, the FDA might come back and pull our approval and say,
go figure out what's the best waveform so we don't mess with it.
These sorts of things happen a lot in the regulatory world with medical devices.
And so we started to say, well, we're going to figure this out.
So in addition to stimulating, so we can stimulate the vagus,
we can stimulate the trigeminal, which is also useful for a bunch of different things,
we would record.
We would record the nerve upstream from the stimulus or downstream,
depending on which way you're interested in having the signal propagate.
We would record the signal in the nerve, and then using just simple signal processing,
we could break it up into different fiber types
because different fibers have different diameters.
Different diameters correspond to different velocities.
Different velocities means that if you record at some distance,
your signals are not going to be separated in space or in time,
depending on how you plot it, right?
So they're going to be separated, and then we can tease them apart
with A fibers being large motor fibers, C fibers being small sensory fibers
going away from the brain, and B fibers being something in the middle.
And then we can start to say, well, what is the maximum amount of activation
you can get at each of these fiber types?
And then which fibers do you want to activate,
and what percent of that maximal activation do you want to achieve?
And so we can determine what the stimulus is,
and we can update the stimulus waveform to give you the amount of activation that we want.
And if we find out that, say, 70% maximal activation of A fibers
is what gives you seizure suppression, then why stimulate B and C fibers?
Those go down to your gut and your heart.
You don't want to be stimulating those.
Actually, it's the other way around. Sorry.
If you want to stimulate C fibers, A and B fibers go down to your heart and your gut.
They change your heart rate. They change your digestive patterns.
You don't want to be activating those.
Why don't we focus on the ones we want, meaning we have less current, less dosing,
and a better longevity for the device,
and simultaneously a better therapeutic efficacy.
Moreover, when we say, all right, we're going to do 60%, say, 60% A fiber activation
and 45% C fiber, which works pretty well,
we're now going to say that that level of therapeutic activation
is what leads to the reduction in seizures.
So we're not going to dial it up until you twitch.
We're just going to give you 60%, and you're not going to feel it.
But you don't need more because more doesn't help you.
And when we implant this thing, we're going to set it to 60%.
But as you age, as you gain weight, as you dehydrate or rehydrate,
as your electrode becomes encapsulated with a glial scar,
it's automatically going to update the stimulus parameters
because it doesn't care what the stimulus is within, say, parameters.
It just wants to achieve that same amount of neural activation
which leads to the therapeutic effect.
And that actually works out really well.
And it doesn't just work out for epilepsy.
I apologize for the print size, but there's so many things that it works out well
that I wanted to stuff them all in one slide, and here they are.
It looks better in a big poster.
But we've been trying this out in epilepsy.
We've been doing it in alcoholism where we have alcoholic rats
that we can stimulate and get them to stop drinking.
We're using this in depression.
We're using this in gastroparesis.
We hear a lot about diabetes, and I've avoided diabetes for years
because I wanted to stay focused on the neural world.
Well, it turns out one of the complications of diabetes
is that the peripheral neuropathy that affects your extremities
also affects your intestinal tract.
And so patients that eat, the food goes into the stomach,
and then it's supposed to get pushed out by the muscles
into the small intestine and then down.
That doesn't happen. It sits in your stomach.
And then after a while, your stomach, your body does what it knows to do
when things are just sitting there, pushes it out this way.
And so you go into uncontrolling bursts of vomiting
that happen every 3 or 4 or 5 minutes for 3 or 4 days,
and this happens once a month.
10,000 hospitalizations every year in the United States is a primary cause,
and 130,000 hospitalizations every year is a secondary cause.
And that's 50% more than 5 years ago.
So it's a bad problem, and it's getting rapidly worse.
And this helps that also.
We can also add a second loop.
So we have the first loop, which is to make sure that we have
the same amount of excitation in the body that we want,
and that we maintain that.
The second loop is that we stimulate only when it's necessary.
So, for example, in epilepsy, we can track a third-party biomarker,
like, for example, seizure onset,
and we can stimulate only when a seizure is about to begin.
Why stimulate all the time?
You're not having seizures all the time.
You're having a seizure once every few days.
If you're really bad, you're having 3 or 4 a day,
but that's still not all the time.
Stimulate only when you need to.
Stimulate only the amount that you need to.
And you minimize the dose, and that means that over time
the patient is going to become desensitized less
than they would otherwise, or not at all.
Another area where we're working on stimulation
is a new field called optogenetics.
And optogenetics is this idea that you can take the genetic code,
originally from algae, actually, that encodes for ion channels.
So you have ion channels in your body that are voltage-sensitive.
You have ion channels in your body that are ligand-gated
or chemically sensitive.
They respond to neurotransmitter concentrations, for example.
There's a third kind of ion channel in algae that responds to light.
And so what you can do is you can take the genetic code
for that ion channel and put it in a mammalian neuron
so that the neurons that get this opsin,
and there's different kinds, there's excitatory ones,
there's inhibitory ones, channelrhodopsin, halorhodopsin,
but the neurons that get this opsin now become light-sensitive.
And the reason this is interesting is because you can make
a very small spot of light on a brain.
So if you electrically stimulate, you electrically excite a volume
of tissue around your electrode tip.
If you do it with an optical fiber, it can be just a spot,
just a few microns in diameter.
And even better, you can separate the viral vector you use
to deliver this opsin so that you have a viral vector
that encodes for the enzyme that brings the opsin into the membrane
and a viral vector that encodes for the opsin itself.
You can put one in over here, one in over there,
and then only the neurons that go from here to here
will get both viral vectors that they need,
and only those neurons will become light-sensitive.
So when you now stimulate in the brain,
you can selectively stimulate a specific tract of fibers
going from one place to another
in isolation of everything else around it.
So it allows you to identify the circuits
that you're interested in activating.
This isn't something you're probably going to do
in a human clinical setting, but in an animal research setting,
it allows you to understand the mechanisms
behind the therapy you want to deliver
so you can make your electrical implant accurately
to deliver that therapy.
So it helps us understand how to better design
our conventional medical devices.
So this is the state of the art.
It's a large laser with a cable on an optical fiber
and a large head mount on a rat.
You can see that it's rather cumbersome.
So what we wanted to do is we wanted to make something smaller.
So we made something that... here it is on a quarter.
It's got a microcontroller.
It's wirelessly powered and so on.
And if we use a brain slice that's optogenetic
and we give it blue light every time you see a little blue dot,
that's a spike of blue light,
then that activates the excitatory channel rhodopsin
and the neuron fires, an action potential.
If we give it inhibitory yellow light,
then it activates the halo rhodopsins
and a neuron that would ordinarily be firing
becomes silent whenever we're stimulating.
So it works quite well.
But that's for a brain slice.
What we really want to be able to do
is we can modify behavior in a full animal,
a freely behaving animal.
So we take these devices and we put it on a mouse
and he's running around.
When the blue light goes on, we begin to stimulate
and he begins to go in left-hand circles.
And whether he likes it or not,
whether he wants to or not,
we can make him move in left-hand circles
because we're stimulating the motor cortex.
We did this in a couple of different animals
and it worked quite well.
We moved from there to doing this
in trying to modulate behavior through reward mechanisms.
And so you can do condition-place preference
or condition-place aversion
where you stimulate the reward centers of the brain
to promote preference
or the fear centers of the brain to promote fear.
There he goes in left-hand circles again.
And you can tell he doesn't want to.
So to promote fear...
I know it's a little sadistic, but...
to promote fear.
And so you can get a mouse that would ordinarily
want to be in a dark room versus a light room
to spend all of his time in the light room,
either because you're conditioning him to like it
or because you're conditioning him to dislike the darkness.
And that's nice,
because it shows you that these circuits,
that it's not just an individual neuron,
but it's an entire neural circuit
and it's a behavioral output.
You can make animals make different choices.
You can take the device out,
remove it entirely,
and the animal will continue to prefer things
it should not prefer, being in the light,
because it's been conditioned.
Its behavior has been changed fundamentally.
And we can start to combine this
with optogenetic stimulation, OGS,
with the autonomous neural control, ANC,
so that you can do this in a nerve
using an optogenetic nerve cuff.
And that's very nice, because now instead of...
So if you go in and you electrically stimulate a nerve,
you're going to recruit the large fibers first
and the small fibers last,
because that's how electricity works, right?
But if you go in there in the body,
when it recruits nerve fibers,
it recruits small fibers first.
Large fibers last.
When you do optogenetics, it does it in the same fashion.
And so using optogenetics,
we have a more physiological way of doing nerve fiber recruitment.
And you can couple that
with the autonomous neural control,
and that becomes quite interesting in its own right.
So those are two of the
stimulating modalities
that we're involved with.
On the networking side,
we try to do two things.
One is we try to power these devices.
So if we're going to make small implantable wireless devices,
we need to figure out a way to deliver power to them.
And batteries won't do,
because they become cumbersome,
and they run out of charge and need to be replaced.
So we need to figure out a way of powering,
and there's lots of different ways to do this,
but what we do is we do far-field electromagnetic coupling.
So you have ambient electromagnetic energy
in the air.
That's some amount of what you can get,
but really you need more than that.
So you can create a signal source,
say a 1 watt, 2 gigahertz,
2.4 gigahertz transmitter.
You can put it in your bedside table,
and then whenever you lie down at night,
it'll drop that 2.4 gigahertz 1 watt signal.
And instead of demodulating it
like you would with your Wi-Fi
to extract the data for your internet connection,
what it'll do is it'll just rectify that incoming signal
and store it in a capacitor.
And not just any capacitor,
but a super capacitor.
More on that in a little bit.
So there's ways of designing these circuits
so that you can optimize the efficiency
with which you capture power
and with which you distribute power afterwards.
And we spend a fair amount of time
looking at frequencies for this purpose.
The frequency selection is non-trivial,
so we spend a fair amount of time
looking at frequencies.
When you think about an implantable device,
you'd think, well, the lower the frequency, the better.
Because the lower the frequency,
the less attenuation you're going to get through tissue.
And that's true.
But then the lower the frequency,
the bigger the wavelength,
and that means the bigger the antenna.
I learned that from Dr. Bernhard
in the back years and years ago.
So there's trade-off.
If you have a fixed size that you have to work with
with your device, say a centimeter,
well, then a big wavelength is going to force you
into an electrically small antenna,
which is challenging, though perhaps not insurmountable.
At the same time, a low wavelength
is going to give you the advantage
of the tissue attenuation.
Add to that a third factor,
which is that the rectifier
for that power is on-chip,
and rectifiers work more efficiently
at low frequencies than high frequencies.
So you have to put all these together
to create a figure of merit that tells you
where is the optimal frequency to operate in
for a particular device of a particular size
in a particular part of the body.
And most of the time, we go to 2.4 gigahertz.
Lately, we've been forcing some things
at 900 megahertz, and it's been a little bit challenging.
A few days ago, we were talking
about building a 60 gigahertz device.
And so there's lots of different frequencies,
and the optimal point occurs differently.
It occurred to me that throwing up
an antenna in front of my old advisor
is a precarious thing to do.
But I will say that this is a meandering dipole,
and that's all I will say about it.
But we do a fair amount on antenna design,
which we don't know how to do well.
And one of the reasons why I was excited
about coming here for this seminar
is because it's an opportunity for me
to bring Henry, who is a student in my lab
in the back, who does antenna work,
and introduce him to my old advisor
and start to work together.
So full circle. It's kind of nice.
Anyway, so we do some antenna design work.
These are efficiencies that we were
looking at a year ago.
More recently, Henry's been able to get
minus 6 dBi efficiency
on a 900 megahertz antenna of these dimensions,
which he's very proud of, and he can show you.
The way that we demonstrate this
when we bring people into the lab
is we take these antennas, we couple them
with a rectifier, with a well-matched rectifier circuit,
like the one shown here, and then
we hook it up with an LED, and we drive that LED.
And you can do it for different sorts of devices.
So from the very small,
like a
intraocular pressure sensor
for glaucoma, there's the LED lighting up.
A larger cardiovascular sensor,
there's the LED lighting up. Now further away,
a TMR sensor for
EMG recording, further away
still, and then if you can cheat with the antenna,
make them very long and flexible under the skin,
well, you can power them from
25 feet away in the lab with a 1-watt
signal source. So
it all depends on what sizes you can
play with, what materials, what form factors.
On the telemetry side,
we began by looking
at, if you have an implantable device
and you need to transmit, let's say that
you're recording data at 10 kHz, right?
And so you're transmitting 10 kHz
with, say, 10-bit resolution to keep the numbers
simple, so you've got 100 kbps
that you're going to be transmitting. Your
transmitter is going to use
99% of the power consumption
of your device. That's just the way it
goes. Transmitters use about a milliwatt, you can do
everything else for tens of microwatts, and
so the transmitter completely
overwhelms your device power consumption.
And that power is coming from that wireless coupling,
which is not very efficient, right? So what
do you do? Well, the first thing that
we started to do is to say, well, why have the transmitter
on all the time? If you can make a quick
turn-on and turn-off transmitter, then have it
turn on, transmit a data point, turn off,
and then wait until
a few hundred microseconds
later when it's got the next sample. Or
alternatively, if you don't need to be real-time, have
it accumulate data for a few seconds, turn
on, transmit all in one burst, and then
turn off again. So if you have multiple
devices, you have some sort of time
division multiplexing scheme in place
where you can have these devices talking
intermittently or spoken when spoken to
from an external base station. So that's
one implementation that we have. But another
one, we had this idea
that we're wirelessly powering
these devices, right? So we have an
incoming RF signal.
And all the energy to transmit comes
mostly from generating
this onboard RF signal that you're
going to transmit outwards. But we already got
an RF signal that's coming in. So why
not take that incoming RF signal,
put it in a flip-flop, divide it
by two, and retransmit it
back out with a modulation.
And so we started talking about this last
year. We designed it on an integrated
circuit. We fabricated the integrated circuit.
It came back three weeks ago. We've
tested it. It works.
And so we've now gone from a
transmitter with a one milliwatt power consumption
to an equivalent transmitter
with a 97
microwatt power consumption. So that's
an order of magnitude reduction. It's a big deal.
It means your battery will last 10 times longer
or be 10 times smaller. And then
we also designed the receivers, of course, that
go on the outside and interface with the implanted devices.
We did this all on an iPad
because it was a nice interface that you could carry around
to demos. But we've got
various different versions
working on it.
The power storage is another area where we
spend a lot of time. So the batteries
have this problem that they're
not rechargeable, or that if they are rechargeable,
there's a finite number of times you can recharge them.
And even then, charging a battery
is tricky, especially inside the body.
If you overcharge it or if you don't charge it at the right
speed and so on, you can get
the battery to do bad things,
overheat, explode, and so on.
So a capacitor would be much better, but capacitors
don't have the energy capacity
of batteries. So what we wanted to do
was to come up with a better power
storage design. And this is what we came up with.
This is a photograph of a supercapacitor.
We have a postdoc from China. His name is
Cuizhou Meng, who came over.
And this supercapacitor, you can
see here, it's got two contact pads.
It's 700x700 microns.
It's got a thickness of 50 microns
and a capacitance of 1 millifarad.
So that's a lot
of capacitance in a very small volume.
The catch, of course, there's two catches.
One is that it's got a very
high effective series resistance.
So if you try to pull any kind of current out of it,
you'd lose a lot of voltage just in your
capacitor. The other catch is that the
breakdown voltage is about 0.8 volts.
So I talked about those circuits that we designed earlier
for power management. A lot of what
those circuits do is
they boost the 0.8 volts
or the 0.1 volt as it discharges.
They boost it up to the 1.8 volts
we need to run our integrated circuits off of.
And at the same time, they
take the charge on the supercapacitor
and maintain that charge
at the right voltage on a conventional capacitor.
So when we have a high current spike,
we can pull it from the conventional capacitor,
which has a low effective series resistance,
and not from the supercapacitor.
So we can optimize efficiency by designing the circuits
intelligently around that.
So that's how we address the wireless component,
or two of the ways in which
we address the wireless component.
The next piece is the miniaturization,
and a lot of effort goes into this.
The first approach, obviously, with miniaturization
is by putting all of the intelligence of these devices
on an integrated circuit.
This is our most recent tape-out run.
Are there any circuit designers here in the room?
No circuit designers here in the room?
Or are you just shy?
All right, she's shy.
All right.
So, okay,
for the circuit designers in the room,
you'll know what this is.
For everybody else, when you get a chip inside your phone,
when you watch, well, it's not Steve Jobs anymore,
but when you watch whoever it is, the new guy,
giving a talk for the new iPhone,
or whatever, right?
Those chips are custom integrated circuits.
That's what we do on a much smaller scale.
We're making custom integrated circuits
for medical applications.
And so when we come up with a design,
we have multiple different projects,
and we come up with lots of different chips.
Each one of these little squares
is a different chip for a different application.
And this is our recent submission.
This is where that new transmitter is on, and so on.
And so, for example, we have a chip on here for glaucoma.
It's 700 by 710 microns,
intraocular pressure when coupled with
the pressure sensor I showed you earlier,
and transmit that out of the body.
We have a chip for seizure monitoring
that allows you to record
electrical activity in the brain
and transmit that out,
and transmit detected seizure events
early in their onset,
early enough that we might be able to intervene
in the seizure before the seizure propagates.
We have a chip that allows you to monitor
pressure continuously in the heart
so that you have an early warning system
for heart failure patients of heart failure.
We have a chip that allows you to record
electrical activity in the muscle
so that you can have patients
that have been amputated, for example,
where you can re-innervate the muscle,
more on that in a little bit,
and you can record that muscle activity
as a proxy for brain activity
and use that to control a robotic limb.
And all of this goes back to the original theme,
which is that a single patient
will not have a single pathology
over the course of their life.
They may have glaucoma,
and at a different point, which is the second
cause of blindness in the world,
they may also be epileptic.
And so you don't need
to treat one in isolation of the other.
You can create a device for one,
and another one for the epilepsy,
and another one for the congestive heart failure,
and another one for the neuromuscular condition,
like Lou Gehrig's disease
or multiple sclerosis, or for the prosthetic arm
that they need if they're a warfighter
coming back from Iraq or Afghanistan
with an amputation. And all of these
devices can be networked together
into a network of
wireless devices that treat the patient
as a whole, rather than as a collection
of distinct and unrelated
pathologies.
So the vision here is that we
partner up with NEADS.
NEADS is an initiative at Purdue,
a nano-engineered electronic
device simulation, which
takes nano-devices that
are available, resonators, biosensors,
a number of different devices, and
models them using
tools like SPICE in the same
way that we model
transistors, or resistors,
or diodes, or other components
which we put on integrated circuits.
And that allows us to design
a full biological
implant, not just
by designing the chip and then separately
designing the sensor and the actuators
that I've shown you and then trying to glue them together
in one package, but rather by
designing all of that functionality
into a single monolithic integrated
circuit. So you can design
it all together, you can then fabricate it all
together, and we've actually started to do this already.
So we've got some devices where you take a
transistor and you coat the gate with an
analyte that responds to a particular
biomarker, and that modulates
the voltage at the gate the same way that
you would ordinarily, except it does
it with a concentration of a particular
protein that you're interested in. And that's a
reversible reaction, and so you can use that
to track protein concentrations in the body.
You also have a device that we just
fabricated, it's a nanoresonator,
and in the nanoresonator
you space dope
channels at a particular distance, you
create an acoustic resonant chamber
on the silicon, which you then
pattern the metal over it to create a resonant
chamber, and you can get high
quality factor resonances. So
the best resonator that you can make
on a standard CMOS process is an
inductor, it's got a quality factor of about 10.
That's the ratio of the energy used to the energy
lost. These have quality factors
of over a thousand, measured
quality factors. So it's not, it's
a two order magnitude jump
after 20 years of
no jump at all, right? So
it's a big deal, and these are nanostructures
which we can now model through
needs, into our systems, fabricate
and integrate. And we
can then package them, right? And I have this
one here, if anybody wants to look, I've got this one
actually here, why don't I pass it around.
So this is a device
we made for muscle activity sensing
and
it
includes the microchip that we
fabricated that I showed you earlier, it includes the
supercapacitors that I showed you earlier, it's
got two contact points and an antenna.
And I use the word antenna loosely because
it looks like an antenna, but it doesn't actually radiate.
Ordinarily I wouldn't say that, but
present company, I felt like I should.
Anyway, my student
is blushing and contemplating leaving my lab.
Anyway, so this is the device.
The package is really cool, that's what I wanted to show you.
We can take these and we can
make them smaller, especially as we start to get rid
of some of these off-chip components and bring them
more and more and more on-chip.
So we can make them really small. Here's
one that's 2mm and the board
itself is only about 800 microns long,
the rest is a loop antenna which sort of
radiates. We can make them
small enough to go inside of a mouse
and not just anywhere on a mouse, we can make them small
enough to go inside the eye of a mouse.
So this is really,
really small.
This is the capacitive pressure sensor I showed you
earlier, fabricating in collaboration with
Bill Chappell at Purdue. There's the loop
antenna and that's a complete
wirelessly powered, wirelessly
transmitting, half millimeter mercury
resolution intraocular pressure
sensor in a mouse eye.
It's small.
We can make them smaller.
So here's the device.
There's a penny, there's a
Lincoln Memorial on the back of a penny
and as you zoom in to Lincoln you'll see next to his
head he's got a microchip and
off of that microchip we've got a wire
which doubles as an antenna
and an electrode so you can record using
it or alternatively you can coat the tip
of it with a biomarker and use it
for biomonitoring and
telemetry. I tried to coin
this one, I tried to call it an antectrode.
I thought that was really cool. Somebody told me
I needed to come up with a new word for this
and I came up with that word and I used it
and I put it on posters
that I didn't take.
But it does work.
So you
start to think about these sorts of science
skills, especially when you can incorporate all
these different technologies on a single system
and now you start to see that using that
magnetic insertion system I showed you earlier
we can inject these anywhere on the body
and we can inject them in a network fashion.
So if
we had the right sorts of capabilities
which we do to some extent in the partnerships
that we have at Purdue but we're growing
then you can envision that we'd be
able to drive forward this area
of personalized medicine around
distributed physical networks
of wireless nanoscale sensors and actuators
and so
we're hiring.
And
for all you future faculty
I'm not
poaching existing faculty
if you are an existing
faculty member come see me
but for all the future faculty
we're talking about nanobiosensors.
We do a lot of modeling with nanobiosensors
but we don't do a lot of fabrication
so we're dramatically increasing
the fabrication facilities that we have at Purdue
and if you know people who aren't in this room who might
be interested please feel free to spread the word
and give them my contact information.
A lot of what we do right now is electrical
sensing and mechanical sensing.
We're moving towards hydrogels
but those aren't very nano but if we start
getting into magnetic sensing and particularly enzymatic
based sensing like what I was showing you with the coding
of the gates of the transistors then I think
we can start to do some really interesting things
in terms of this
nanobiosensors
packaging. Everything that the
FDA from a regulatory standpoint
cares about is hermeticity. They want
to make sure that the
device and what's inside it is not going to hurt
the body which is fun
because the devices almost never hurt the body.
It's actually the body that hurts the device.
You take electronics and you put
it in a warm salt water solution
for the next 80 years. That's not
good for it. It's the worst possible thing you can
do to an electronic board.
We worry about the hermeticity
because we want to protect our electronics
not the patient.
We talk about ceramic packages
and titanium packages and all sorts of different
materials but these packages aren't
very interesting from a functionality
standpoint in the sense that it's very
difficult to incorporate
the sorts of microelectromechanical
system structures that we want to
within those packages that they're rigid
not flexible. So we want to have
these features and that means we need to
move in the polymer direction and polymers
are by definition almost
not hermetic. They're near hermetic
so trying to make these package
materials in ways that are interesting
incorporating technologies that are exciting
in ways that are hermetic enough
for the amount of time that we need
this is a major direction thrust
as we move forward. Energy harvesting is
the last one I wanted to talk about.
So we do a lot of inductive coupling.
This is the only one that's approved by the FDA right now.
I talked to you a little bit about RF
powering. You can do it with high intensity
focused ultrasound or optical
but these really don't have a lot of patient
compliance and what I mean by patient compliance
is that they require the patient
to do something. So if a patient has
to take an external charger and hold it up
to their implanted device to charge it
you're almost guaranteed it's not
going to happen. It's like the glaucoma
story where the patients get drops. The drops
will take care of the glaucoma and yet
the patients almost always go blind.
It's because they don't take their drops
and when you ask them they lie to you and tell you
that they did. Patient
compliance is the bane of the medical
world. What you need is a
device that is patient proof.
So you need to get away from devices
that require you to do something
and towards a device like RF powering
which is more or less somewhat patient
location independent that you might be able to
put on your bedside table and as long as you get into bed every night
it'll charge your device
and preferably even towards a device
which harnesses the thermal and mechanical energies
that exist already in
the patient and uses
those to power the implant. So we have three new
faculty slots in these three new areas
and with them the sense
is that we can really begin to enable
these wireless networks. So here's an example.
This is Claudia. She's a patient
up at the Rehabilitation Institute of Chicago.
This is through the partnership with the DARPA grant
that I'm the PI on and she has
a series of sensors located
on her left breast. So
when she lost her left arm, not in combat
motorcycle accident, when she
lost her left arm she
well, it transformed her life
as you can imagine, but what the doctors noticed
was that the nerve that still went down the arm
was still there. And her pectoral and
lateral muscles, they're still there, they're just useless
because there's no arm to move around.
So why not take the nerve that goes to those muscles
and remove it, de-innervate the muscle,
take the nerve that used to go down the arm
connect it to the muscle, over
time it'll regrow into the muscle and re-innervate
the muscle, and then when
you tell Claudia to open her hand or
extend her arm, these muscles
contract in unusual ways.
We can use the device that's getting passed
around, use 10, 12,
16 of them in the muscle
positioned throughout the muscle body
to extract the electromyography,
the muscle electrical activity
that we can then transmit wirelessly
to a prosthetic arm, allowing her
to control a robotic arm
with more degrees
of freedom and more sensitivity, more accuracy
than anything that's ever
been done out there. And this is not
I can say things like better than anything
ever because it's not my work, it's theirs.
It's the rehabilitation's work.
What we're doing is we're looking
at that device and saying no patient
is ever going to use that because those electrodes
first of all, they're positioned on the skin
and when the muscle contracts it moves
with respect to the skin so your signal source is moving
with respect to your amplifier and that does bad things
to your algorithms. Second
of all, every time you go to bed you take the electrodes off,
you put them back on in the morning, they're in different locations,
you've got to reprogram the whole system. It's a big
cumbersome pain in the neck and that means patient
compliance goes to zero.
So you actually take one of these arms, the first one
cost $100 million to develop.
Only DARPA would spend $100 million
that way. And it was really cool
but it was just the first one, right, and it was buggy.
DECA made a better one, that's the one you just
saw. You take
these arms, you give it to the patient, they think
wow, this is really cool. They show all their friends,
they go home, they go to bed, they wake up in the morning, they
throw it in the closet, they grab their hook
and they put their hook on because the hook is
lightweight and it works.
You don't have to do anything fancy.
Now if we use these electrode arrays, we can
get these signals reliably, repeatably
so all she's got to do is just put the arm on, just
like a hook, and it'll automatically work
because it's already been calibrated
and that only has to happen once. So those
are the sorts of things, right, especially when you start
a couple of them with other devices and other conditions
that we're
talking about here. So this is
not just me, of course, it's mostly
not me, it's the students, it's the partnerships,
the collaborations, the sponsors,
the all-important sponsors primarily
for the work I've talked about today is
DARPA, CyberOnyx, and
Howard Hughes Medical Institute
and a number of graduate students,
postdocs, research scientists, technical
staff. They do
all of the real work.
Thank you very much.
Yeah?
How strong of a magnetic field do you use to pull the electrodes down?
That's a good question.
I don't know.
So what we did is, so you know what a
rail gun is? Yeah. Okay.
So basically what we did is we built a rail gun.
Initially what we wanted to do is try to figure out
whether or not we could even get these things to move.
So we made the electrodes a little bigger, 100 microns
across because 20 microns was a little
challenging, there's not a lot of material there.
And then we took a coil we made, we just wound
and wound and wound and wound, and then we got a
capacitor, we charged it up to 400 volts
and we shorted it across that coil.
Blue sparks and all that good stuff
and we shorted it out between the turns and everything else.
But the electrode came out and embedded itself
in a door on the other end of the lab.
So that was fun, right?
And so it's not often you get to actually
build guns in the lab and call it
medicine. But anyway,
so we did that. Then we wanted,
somebody wanted to see if we could do a craniotomy
and so we took an animal,
it was euthanized, it was already dead.
But we just took the electrode
and we were actually able to penetrate through the skin
through the bone and into the brain and bury
in the base of the skull. So we had
too much juice. And so what we
did then is we calibrated a curve
for different electrode thicknesses,
different voltage amounts, different current
or numbers of turns and coil diameters
and so on. And so we had sort of
a tuning curve for what depth you wanted to get
you'd apply this much juice versus
the amount of juice.
The next step was to try to figure out if this
was all really being done as efficiently as possible.
So I have a new PhD student that just started
a couple months ago. And he's actually started
going back to Maxwell's
equations and trying to figure out, alright, how does this
actually all work out?
And it's actually a lot of fun because it's been
a while since I've gone through these equations.
Sorry. But
I see this and
so he looks at this and he says, well actually
the electrode initially is at a stop
and so you don't want to apply the maximal force
there because a lot of it's going to go wasted. You want to accelerate it
slowly and you want to have increasing field as the electrode moves.
So you stack the coils so there's a few turns
up here and high turns up here and he comes up
with these pyramidal structures that actually
work really well. The next problem
that we have, so we don't know
what the magnitude of the magnetic
field is. It's
mostly trial and error but now we're just starting
to try to figure out what the theory behind it is.
What's interesting is that
he came up with an idea for having a
ferromagnetic holder
with a
DC magnet at the tip that holds a
non-magnetic electrode. And so you
can put this in the same barrel and fire it
and then just at the critical moment
you have a DC magnet around the outside
and it opens the holder and so
the non-magnetic electrode will proceed.
It's more of a javelin throw than a magnetic field
pulling it in. But it'll proceed into the tissue
and what you're left with is a
non-magnetic electrode which potentially
you could then use to measure single
electrical activity coupled with
fMRI activity in an
MRI environment.
Yeah, sure.
When you were talking about
how you were powering devices, you talked about
how one of the frequencies
you used was the 2.4
gigahertz. Now that's a pretty common
frequency, right? Using cordless
phones, etc. And if you're also
trying to transmit out the data on there
are you worried that
is it possible that people moving
about, entering areas with those
fields is going to interfere
with the monitoring of it? So if you're trying
to make a device that's going to see clinical adoption
and the clinical impact is what
we're all about
then you need to operate in a frequency band that
the FCC is going to let you stay in.
And those are ISM bands.
Industrial Scientific Medical bands.
The reason there's a lot of stuff at 2.4 gigahertz
is because it's an ISM band which means you can
do just about anything there.
So you've got Bluetooth and you've got WiFi and you've got
everything for us.
Originally what we've
done lots of different variations. One time
the chip that we did a year ago, we had
an OOK transmitter which is on-off
keying. So the transmitter's on when it's
a digital 1 and it's off when it's a digital 0.
And it was at 2.4
gigahertz. And we thought, oh, great.
On-off means that it's off half the time on average
and that means that you're saving half the power.
And that's kind of clever and
it's not a very complicated modulation scheme. It's easy
to modulate. It's perfect, right? Why would you do anything
else? So we fabricated it and the damn thing
didn't work. And the reason it didn't work
is because any time the receiver
sees a 2.4 gigahertz
pulse, it registers a 1. But if you're
in a room in which somebody's cell phone is downloading
data and you've got somebody's Bluetooth headphones
going off all the time, every time each of
those devices sends out a spike, the receiver
picks it up as a 1. So we had huge
error rates which drove our data transmission
rates down because we had to retransmit everything
all the time. So you do have
to worry about it. More recently
we've done this divide by 2 scheme which is nice
because you can power it at say
2.4 or 900 megahertz
and then you retransmit at half the frequency
450 which is out of band. But that's
okay because these medical devices are
so low power in their transmission
that they're actually below
so FCC has guidelines for intentional radiators
and unintentional radiators. You're
transmitting below the intentional
radiator, you can get away with it.
Which is one of the things that we're doing.
You can also shift things a little bit so you can
try to be in the mixed band on the transmit
which is 430 megahertz and a 900
megahertz on the receive.
There's lots of games that you can play.
But yes, it is generally
a bad idea to receive and transmit at the
same frequency.
But you can do it. The intraocular pressure sensor
that I showed you does just that.
That's an FSK
transmitter though so that's always on.
It's easier to pick up the noise.
Yeah.
So the LED will
light up with as little as a few
microwatts. It varies though of course. You saw the
LEDs go super bright and then get dimmer.
So you can put milliwatts in.
They're rated at milliwatts but you can light them up
with just a few microwatts.
Generally speaking what we need for our
medical devices is, so we're
looking at a peak power. It used to be a milliwatt.
And so we were thinking if it was a milliwatt
we could have it on one tenth
of one percent of the time. So
0.1 percent of the time because
the rest of the time it's off between samples.
So you can get the ramp up and ramp down
which we haven't been able to.
Then you could get the average power consumption
of a microwatt. The average power consumption
is what you actually care about.
With a new transmitter we've got
our quiescent or our peak power consumption
down by a factor of ten.
So that helps quite a bit.
And it means that we only have to reduce,
we only have to be on one percent
of the time instead of having to be on one tenth
of one percent of the time. And now I think it's more realistic
that we'll get an average power consumption of a
microwatt. But that's the goal, is one
microwatt average power consumption which is a little
less than what we need to light up those LEDs.
I mean what's the peak power of the
transmitter? The peak power of the
transmitter is 96 microwatts.
Microwatts?
Why does my glaucoma
device need to dock to my epilepsy
device and my heart failure device?
Because let's say
that you have a glaucoma and you have a cardiac
condition. Pressure in
the body is linked together. So
variations in pressure in one area are linked
to variations in pressure in the other. And reducing
pressure in one area can track pressure
in another or increase it in another. So they're
linked. Drugs that you take
to affect your blood pressure
might increase the pressure inside of your
eye. So these diseases
can be linked. Particularly if you start to think
about multiple different conditions inside of the brain.
So if you have a traumatic brain injury
or you have an addiction but you also have epilepsy,
you're talking about neural recording and
stimulation from multiple regions in the brain.
There's no reason to have one device that
does multiple regions for one condition
and another device that does multiple regions for
a different condition. Why not have a network
of sensing electrodes and a network of stimulating
electrodes that can address multiple
different pathologies now?
And as you age
and you develop Alzheimer's, we can start looking
at doing things there as well.
So it's a way of centralizing
all these different things. I have a friend
of mine with whom I work with on the glaucoma
who had high blood pressure.
So he started taking medication for his
high blood pressure and it's
irreparably damaged his neuromuscular
junctions. So that he now
spends three hours every day stretching out
his muscles because he wakes up in a cramped
fetal position every morning. And that's a
side effect of the cardiac medication.
So the downside of
treating these pathologies individually is that
patients with one end up exacerbating
the others because of the side effects
of the drugs they're taking for that one condition.
The upside of treating
the patient as a whole and networking
these devices is that we can try to
get around those sorts of
problems, which are surprisingly common.
That's a good question.
So how do you see that changing the entire
field of medicine? Because it's going to have an impact
on the specialists
in particular. So how
do you see that interaction between them changing?
So let's take a condition
like epilepsy with which we
spend a fair amount of time.
Even leaving aside the networking
properties. Epilepsy affects
1% of the population.
Of that 1%, that's 3 million Americans.
It's 1% worldwide,
but just in the United States, it's 3 million Americans
roughly. 30 to
40% don't respond to drugs.
30 to 40%
respond to drugs, but the side effects are so
severe that they end up not taking their
drugs. The end result
is you get about 54,000
deaths from epilepsy every year in the United
States. That's more than breast cancer.
That's because drugs don't work.
They work for some people, but for a
huge percentage, about 2 million Americans,
they don't work or they don't work well enough to make it
worthwhile. These same people
who stop taking the drugs because it's not worth
it living with the side effects
to stop the seizures can often
die from a condition called pseudepsy.
That's an unexplained death from epilepsy
where you go into a persistent seizure and
your cardiac arrest,
respiratory arrest follows.
And it's an unexplained, that's why the
unexplained death. It's kind of terrifying, right?
For a medical acronym. Kills thousands of people
every year.
The current treatments
are not sufficient.
Folks who go on drug treatments, whether
it's for pain management or for epilepsy
or for Parkinson's or whatever, over time
the body becomes desensitized to the
drug. Not to mention that the drug is
distributed throughout the body and you only need it
one place. These approaches
allow you to target therapies so
that you're going right where you want to be going.
You want to deliver the therapy only when you need
to and you avoid chemical
desensitization because there is no chemical
to become desensitized from.
The notion that you can move away
from pharmaceuticals and that you can
move towards a non-drug, a drug
free therapy program
and that you can do this in a way that you're targeting
it to the area that needs it.
These are all revolutionary ideas
and that you can network it together so that you're
looking at the side effects of a particular
treatment in other regions
and you're treating them all as a whole.
All of these are
different ways of looking at medical practice
than what is done right now.
We have quite a bit of support. We have partnerships
with a group called Goodman Campbell Brain and Spine.
It's a merger of all the neurosurgical
groups in Indianapolis, which is now the
largest neurosurgical group in the United States
and one of the best. We partner with them
to do human clinical trials to try these things out
on humans. Our sense
from conversations with the doctors
or with industry is that
there's a willingness
to begin to explore and
commercialize and deploy
these treatment modalities that didn't
exist maybe 10 years ago.
That's partly because the technologies are much further
than they were then and partly because a sense of the
limitations of more and more and more
money at high throughput drug screening
has not yielded increasingly
better results.
Okay, so let's thank
Professor DeRandoki for a very interesting talk.

