High Resolution Brain Machine Interfaces Using Flexible Silicon Electronics
Sep 5, 2023
High Resolution Brain Machine Interfaces Using Flexible Silicon Electronics
Jonathan Viventi, Polytechnic Institute of NYU
Alright, thank you very much for the introduction.
So thank you very much to the students who are organizing this conference.
I think you've done a great job today, everything's running very smoothly.
So let's get a round of applause for them from all the faculty who have sat back.
And so thank you again for the invitation to talk.
We'll share some of the research that we've been doing in collaboration with John Rogers
who's here at the Beckman Institute at the University of Illinois.
And since this is a student conference, you can feel free to interrupt me at any time,
ask questions.
If you want to know more about my career path, trajectory, things I've done, feel free to
ask at any time or come up afterwards.
And so we work on incorporating flexible silicon electronics directly into the electro-tissue
interface.
We want to do this so we can really improve the ability of us to capture information from
the brain and over large areas.
So our motivation for doing this is that impliable devices have evolved tremendously,
but the basic electro-tissue interface to the brain hasn't really changed very much
in 50 years.
And so these are some examples of systems that are currently in use in patients.
These are responsive neurostimulator and vagus nerve stimulator, and these are devices with
pretty impressive computational capabilities, wireless capabilities, long battery life.
But they would really rely on very simple, very crude electro-tissue interface devices.
And so what would we really like to have instead is dense sampling over very large areas of
the brain.
And so if you take a look at some of the currently available state-of-the-art clinical devices,
these feature large electrodes, three or four millimeter disks spaced almost a centimeter
apart.
And so if you think about this from neuroanatomy, underneath each one of these electrodes there's
over 12 million neurons.
And so we're trying to sample all of that activity with just one contact.
And the limiting factor in all of this has been wiring, and we'll talk more about that,
but if you can imagine this bundle of cabling, and this is what we regularly implant in patients
with epilepsy.
And this is to diagnose patients with epilepsy that have failed one or more medications and
are candidates for surgery.
We place these devices on the surface of the brain, we map out epileptic regions, and then
based on this crude set of information, we go back in and cut out a big area of the brain
under the hopes that we can try and treat their seizures.
And these surgeries are pretty ineffective at best, they're sort of 40 to 60% likely
to give you seizure freedom or even a reduction in seizures.
So you're undergoing a pretty invasive surgery for a pretty low chance of surgical success.
So instead we'd like to build better technology to understand what's going on in the brain
in a much finer scale, and try to come up with better ways of treating epilepsy without
removing large chunks of brain.
So some of the other attempts have been made to capture information from very high resolution
from very small regions of brain.
So this is an example of a device that's very popular for research, for research in brain
machine interface.
It's the Utah Electrode Array, and it features about 100 electrodes that cover an area of
4x4 millimeters.
And in context you can see just how little of the brain surface that can really cover.
So it gives you a very nice picture of high resolution activity, but from a very small
area.
And so our group and other groups have looked at using hybrid electrode arrays, and that
is incorporating small micro wire bundles into the surface electrode, just to give us
an idea of what kind of activity is occurring on the surface of the brain in between these
larger electrodes.
These are spaced a millimeter apart.
And so we look at this electrical activity in epilepsy patients, and we found events
we're calling micro seizures.
These are events that occur on a single electrode and not a millimeter away.
And they look just like the clinical seizures we observe, but we don't see them spreading
throughout the brain.
We see other events, micro periodic electroform discharges, again on a small number of these
micro electrodes and not spreading very far apart.
As well as these high frequency oscillations, these are short transient bursts of sinusoidal
activity that are again confined to very small regions of the brain.
So we don't really know what these activities are doing to contribute to large scale clinical
seizures, but we need a technology that will allow us to record all this micro activity
and piece that together.
And so we're looking at ways we can do this.
We'd like to cover an area that's approximately eight by eight centimeters.
This is a six by six centimeter grid, and this patient actually had two of these different
grids implanted.
So we'd like to cover a pretty significant area of the brain, but if we wanted even one
millimeter spacing across the brain using individually wired electrodes, using currently
available technology, we would need 6,400 electrodes and correspondingly 6,400 wires.
So you can imagine this is about 100 something wires, and what would happen if we tried to
increase the wiring density even further?
Because these devices, now after they've been implanted, the surgeon will go back in
and place the skull back on top, the bone fragment of the skull, and they close up the
dura on top of these devices.
And so any additional volume of material we place inside these patients causes more pressure
on the brain, more risk of injury.
So if anything, we'd like to decrease the amount of wiring while simultaneously increasing
the amount of electrical coverage we have.
So the solution to this has been known for a long time.
We need multiplexing.
We need an ability to record from thousands of contacts using a much smaller number of
wires.
And so some groups have looked at this as early as 1986.
They've incorporated silicon transistors directly at the electrode tissue interface, and as
a way of combining signals from many electrodes to a much smaller number of wires.
But all these approaches in the past have been limited by the material they're built
on.
They've all been built on rigid silicon that is inflexible, it can't conform to the surface
of the brain.
So it can make penetrating arrays that can be pushed into the brain, but not ones that
can really conform to the irregular surface.
And so we're interested in building devices that are extremely flexible.
We want to record from the irregular surfaces of the brain, but also get access to areas
that previously have been inaccessible, like inside of sulci, and inside of sulci and in
between the two hemispheres of the brain.
We'd also like to have devices that can sample at about one kilohertz to fully capture the
information that's available.
So I've mentioned high-frequency oscillations, so these are short bursts of high-frequency
activity.
As an example, something happening at 100 hertz, and they happen up to about 500 hertz.
And so we don't yet fully know the physiologic meaning of these oscillations.
We do observe them in epileptic patients as well as normal patients, albeit at a lower
rate than normal patients.
So we think they're a marker of epileptic tissue.
Further, we'd like a device that can measure brain signals with very low noise.
We're looking at microvolts signals, tens of microvolts, hundreds of microvolts.
And also meet all these requirements while consuming a very small amount of power.
So we need this device to, at first pass, needs to consume such a small amount of power
that doesn't heat the brain by more than one degree Celsius.
And of course, the power requirements are even much more strict if we want to run the
device on batteries.
And so when I started this project, as part of my PhD, we met up with John Rogers.
I sort of found the research of John Rogers here, who had figured out a technique for
building flexible and stretchable circuitry using very thin pieces of silicon.
And the idea is that silicon, when we think of silicon wafers, we think of very rigid,
brittle material.
But if you make it extremely thin, it can be very, very flexible.
The same way, this is a great analogy that he uses, that if you have a piece of two-by-four
lumber and a sheet of paper, one is inherently flexible, one is inflexible, but they're
the same material.
So he figured out that if you make the silicon transistors on 100 nanometer thick nanoribbons
of silicon, you can make them flexible and even stretchable.
And so we collaborated to develop the medical applications of this technology.
And so we started by developing these arrays of active electrodes.
And since this is a student conference, I thought I'd throw in a little bit of background.
Everyone's probably seen a very basic circuit diagram.
This is how all these devices work.
There's a signal recorded from each electrode on the surface of the device that's connected
to the gate of a flexible silicon transistor.
And then there's a constant current sink provided off the array that forms a source-follower
amplifier.
So we're really not providing any gain at these devices, just buffering.
We want a nice low output impedance, much lower than the impedance of the electrode
on the brain.
And so that allows us to do multiplexing, as we'll see later.
And so this is a device we're currently building.
This has 360 electrodes arranged in a 20 by 18 pattern.
They have 500 micron spacing between electrodes.
And because we've built those amplifiers, those source-follower buffers and multiplexers
directly into the electrode array, we can record from all of these electrodes with only
39 wires.
So we can take a little bit of look at what's going on inside each one of these cells.
It's the same basic design that gets repeated over and over again.
And it consists of two transistors, two flexible silicon transistors, and of which one of them
is connected through this vertical connection to the surface electrode, the platinum surface
contact, directly to the gate of this buffer transistor.
And there's a second transistor that provides a switch that turns on and off when we enable
this row select signal that runs across the array, and allows that electrode to connect
to a shared column output line.
And so all the electrodes in a given column share the same output wire, and off the array
we provide this constant current buffer.
So just running through this multiplexing with the same idea is we can turn on all the
electrodes in row zero and sample all of their activity, then turn off row zero, cycle to
row one, sample all the electrodes in row one.
If we can do this fast enough, if we can cycle through all the rows in the array,
we can fully reconstruct the original signals that are available across the brain.
So this device we built up in a series of fabrication steps, and it starts off with
a flexible silicon nanoribbons that have been transfer printed using a rubber stamp onto
this polyamide flexible substrate.
So it could be on a number of different flexible materials, and John's group has demonstrated
active devices built on paper, on leather, on stretchable surgical gloves, really any
material they've been able to build circuits on it.
In our case we're using polyamide.
And then we build up the circuit interconnections that form the device in a number of steps
that are on this flexible substrate, and finally we deposit platinum contacts on the surface
for the electrodes.
Now in cross section this device, you can see the silicon located in the middle of the
device, and then numerous interconnect layers, and finally the platinum electrodes on the
surface.
The whole stack right now is about 25 microns thick, so you can think of it as a very flexible
cellophane like material.
So certainly we like to make it much more flexible, we're working on making this thinner,
making this even more flexible, but it's still something that can conform to a little bit
of curvature of the brain.
So we've taken this device and done some preliminary experiments with it.
We've placed this over a cat visual cortex with our collaborators in neuroscience, and
we recorded from anesthetized animals visual evoked responses.
We can also take this device, because it's really flexible, and insert it into the inner
hemispheric fissure.
So this lets us get down into that area and get at some previously inaccessible areas
of the brain that are very hard to record from.
We can also take this device and we actually fold it back on itself around a very thin
sheet of PDMS, of rubber, and make a unique kind of double-sided recording electrode.
So we can record from both sides of this device.
And we've also placed this into that inner hemispheric fissure as a model of what a human
sulcus might be, roughly dimensioned.
And so we can show that we can record from both hemispheres of the brain as well as both
sides of a sulcus.
Do you have to do anything to decouple those two sides?
Yeah, so there is an insulator in between the two, and so each one of those electrodes
is isolated and buffered.
Does the insulator need to wrap that as well?
That's a good point.
If you want to really insulate the two, that would be a good idea to put something to really
shield the back electrodes from the front.
It could be, that could be definitely done.
Although a lot of the contact is directly, the recording is from direct electrical contact
with the brain, so it's like chemical contact.
But that's a good idea.
And so we can do experiments with this device.
So we can record from the visual system.
And so we've done some evoked response, some very basic evoked response experiments where
these are anesthetized animals.
We're showing an eight by eight checkerboard visual stimulus.
And so this is an example of the movie that we show the animals and record their responses.
And so there's a box flashing in one of 64 locations, kind of very basic neuroscience
experiment.
And what we can show from this is that we can pull out features from the data, and this
is the response for the entire electrode, all 360 channels for a box presented in the
upper left-hand corner of the visual field, and this is a box presented in the lower right-hand
corner of the visual field.
And so we can extract out some basic features like power and delay to the peak of the response
and use these to train a classifier that does prediction.
And so we were able to predict that 36% of the locations exactly correct and 66% of them
within one adjacent square.
So not really very optimized, but a demonstration that we're able to record useful signals that
can be used to do decoding.
Not quite as impressive as decoding YouTube movies, but got to start, I guess, somewhere.
So we'll try that one next, though.
No, we're not going to show YouTube movies to cats, I guess, but...
Oh, yeah.
So, cat movies, right.
There's a lot of cat movies on YouTube, so...
So getting back to epilepsy, that was the original purpose for these devices.
We were able to induce seizures in these animals using drugs to try and look at how seizures
propagate throughout the brain.
So this is an example of what clinicians currently look at.
They look at electrode data, a few channels at a time.
They plot it as raw voltage data, and they start trying to compare the waveforms between
these channels.
And so if you or I look at this, these are three randomly selected channels from the
array.
It's hard to see if there's any specific pattern.
Maybe the shapes are a little bit different between those channels.
But with 360 of these channels, this is just a few seconds of data.
This is not really a feasible thing to look through all of these channels.
So instead, we plot this as a movie.
We take the raw voltage data on top here, and then represent the voltage as a color,
and then show the average of all 360 electrodes at the bottom as a correlate to what you would
see with only one electrode in a square centimeter.
So we're seeing a lot more information.
We have 360 channels in the same space as one before.
And when we plotted this seizure, a number of things really struck our eye.
Then we see complicated spatial patterns moving across the electrode array during this seizure.
And so we see waves moving in different directions.
They sort of start and rotate in one direction.
And then the waves sort of at some point decide to change directions and move in a different
way.
And so this whole pattern kind of continues throughout this seizure in a kind of very
robust manner.
And this has been slowed down 20 times.
So now we're looking at this very much slower scale than the original data.
And so we're interested in taking these patterns and trying to characterize them and pull out
information about what is going on.
Are they predictive to seizures, and are they informative to therapy?
And so we looked at some of these patterns.
We tried to pull out automated methods to analyze this data because, as you can see,
this is just a few seconds of data.
And we have hours and hours of these recordings that we can't review like this.
We can't review manually.
We needed automated methods.
And so we started with a very simple analysis method.
This is a basic spike detector.
We're identifying spikes on all of the channels.
This is an example of a movie of one of those spikes.
And so for this spike, we repeated three times.
It appears to be very stereotyped.
Starts in one direct corner of the array and moves to another side of the array.
So from that, we can calculate the delay from any channel to all the other channels, from
the average to all the other channels using cross-covariance.
And we can build a map that reduces that original movie down to just one image that says this
spike moves from here to here.
We can also extract the power in each spike, in each electrode.
And we can use those two sets of features.
We can form PCA on them to reduce from seven or 20 dimensions down a little bit farther
to about 60 dimensions.
And then do clustering.
We wanted to find out if those spatial patterns that we observe, are they repetitive or are
they just all possible capability, all possible patterns occur within the data set?
So really is it one big cluster or are there specific clusters within the data that are
informative?
So we use clustering and then we use a gap statistic to look at the data.
And this isn't our data.
We're trying to find in a statistically significant way where the elbow is as you increase the
number of clusters.
As you increase the number of clusters, the within cluster distance goes down.
But at some point, there's diminishing returns.
And so you might pick that, for example, for this toy set of data that there are two clusters
in the data.
And so we did this with our data set and we found 21 significant clusters in the data.
And this is an example of one of those clusters that really shows that all of these spikes,
no matter where they occur in the recording at different times, they look very similar
to each other, at least somewhat similar to each other, but very different than spikes
that occur in another cluster.
And so these spikes were very stereotyped and looked and repeated this kind of pattern.
And so now we can go back with these labels and look at that original seizure again.
We can look at late with labeled data and we can make up labels for these patterns as
well.
Of course, the algorithm doesn't spit out the descriptive labels and we can get assigned
these descriptive labels.
And it appears the seizure starts with one particular pattern, this plane wave followed
by these three counterclockwise spirals, clockwise spirals, sorry, and then another plane wave
and then this long period of prolonged counterclockwise rotation.
And then finally, it seems to end with this wave moving right to left and then the same
wave repeated over and over again, that purple wave.
So it's just a way of us looking at this raw data, pulling out patterns that are repetitive.
And we can do this across all the data sets, across all the time we have.
And so this is sort of led us down the question of what do we do with this data?
We're interested in comparing these patterns to, now we have all these clusters and all
these labels, what is their relationship between these types of spikes and seizure initiation,
seizure propagation, and ultimately seizure termination.
So we'd like to figure out if there are specific patterns that either precede seizures or end
seizures, can we identify when seizures are going to occur and cause them to either not
occur or to terminate sooner through electrical stimulation.
And so this is what we started to work on now in my group.
We're developing electrode arrays that incorporate multiplex stimulation as well as recording
directly in the array so that we can try and deliver the same stimulus that we identify
from the recordings that might terminate the seizures.
So we can try and deliver this kind of wavefront stimulation to block moving and recurrent
patterns that are moving across the brain.
So to summarize a little bit, the electrodes we've built so far are about 25 microns thick.
They're pretty conformal to the brain.
We have some techniques with silk that we're working on that we've published a little bit
earlier about how to make them even thinner.
Those devices were two and a half microns thick.
So we'd like to make them much thinner, much more flexible.
We're building arrays that have up to 1,024 contacts.
The next devices we're building are even more electrodes with even finer spacing.
We have some other designs that push the spacing down to 250 microns.
We sort of think this might be a good place to sort of match the natural resolution of
the surface of the brain of the information available.
Because we're using this high-performance monocrystalline silicon, we can switch between
all those multiplex channels pretty quickly.
And so we can sample each of the electrodes at up to 10 kilohertz per channel in these
devices.
And so much probably faster than we need right now.
We need about one to two kilohertz.
We've integrated multiplexing and amplification at each electrode.
And so the 1,024-channel array will have 37 wires.
In the future, we're looking to scale this to even more, even thousands of electrodes
and even fewer wires as we try to work towards larger coverage, larger animals, and chronic
implants.
Some of the devices we're building and some of the future work, we're just starting in
my lab.
So I've just had all my students just started.
I've been recruiting students and building the lab and ordering equipment.
And so we're just starting some of these projects to incorporate stimulation at all of the electrodes,
as well as building devices that can be chronically implanted.
All the things that we've shown so far are acute experiments.
So we've exposed the brain, placed the electrode on the brain.
So we're working on how do we implant these chronically over long periods of time.
We're interested in incorporating wireless capabilities.
So we have a sort of mock-up of a concept sketch where we're going to take inflexible
silicon and attach it directly to the subflexible substrate at the periphery, at the areas where
it doesn't need to be that flexible.
And use that inflexible silicon to do analog-digital conversion, to do the control signals, and
ultimately build in the wireless capabilities we need to get this data out of the head.
And finally, we're looking at using these devices for brain-machine interfaces.
We're partnering with a collaborator in neuroscience in the Center for Neuroscience, Bijan Pesaran,
who's working on looking at BCI and improving BCI, looking at LFP and looking at these arrays.
And our hope is that by taking arrays of 1,000 electrodes on the surface of the brain,
we might be able to get out a little bit more information in a more reliable way than using
arrays of 100 penetrating electrodes to make a reliable, long-term neuroprosthetic device.
So I'd like to acknowledge my support.
We have an award from CURE and NYU Wireless.
And thank you for your attention, owned, of course, while I'm recruiting.
So take a look at the website if you're interested for all the students who are graduating.
Thank you very much.
Questions?
Yeah.
Yeah, those are good questions.
So the implant devices today either have non-rechargeable batteries.
So some of them, you design the system to last five years on a battery,
and then you do a surgery to replace the battery.
And for some of those implants, the electrode stays in.
You remove the pulse generator, the whole implant, and you replace that part and reuse the existing electrode.
Other systems are rechargeable.
So there are a device in trials that is rechargeable.
It's implanted here, and the leads are tunneled up to the head.
And then the patient every day puts on a thing that looks like a scarf while they're watching TV,
and that inductively recharges the battery.
So for devices that really have high drain, that's the only way you can achieve what you need to do.
But you're still pretty limited, even with rechargeable batteries, to what you can do in the long term.
The other type of devices that don't have batteries at all, like cochlear implants,
are inductive links that supply all of their power.
Yeah?
How much brain remodeling goes on?
Yeah, that's a really good question.
So some work by Justin Williams at the University of Wisconsin is looking at this.
And so we sort of think that, well, surface electrodes, they don't do anything.
They're pretty inert.
You're not penetrating into the brain.
And you're on top of membranes.
But there is a significant amount of growth of tissue that sort of comes around and encapsulates things.
And we don't think it really affects the signals you're recording,
since you're recording from such large groups of neurons.
You're recording from thousands, tens of thousands, hundreds of thousands of neurons still.
But it's interesting to see how that's going to affect your electrode, actually the brain from your electrode.
And so some of the work he's done is basically putting holes in the electrode,
allowing CSF to circulate.
Because if you have this big plastic sheet, you're impeding the ability for diffusion to occur.
And so we're going to try and look at that.
And we may have to sacrifice some of our resolution to incorporate that circulation into our devices.
That's very good.
We're going to try to look at some of that.
We're going to start doing chronic implants, hopefully in the near future.
So we'll see what happens when we pull out the arrays after a few months, what they look like.
Thanks.
Yeah.
How consistent are those seizure patterns that you were showing, seizure to seizure and subject to subject?
Right.
Or do you expect real-time learning to happen in the device in order to counteract?
Right.
I wish I knew.
That's the big question.
We have a few examples from just a small number of animals,
so we can't really say conclusively how consistent they are between different animals.
They're probably going to be patient-animal specific.
In one example, in one animal, they seem consistent over a few hours in that they have very similar kind of waveforms.
Although within one seizure, there's one example where it progresses from one type of pattern to another to another.
So those are all good questions, and hopefully we'll figure it out.
What is the cost differentiation between your new product and the standard current product?
Right.
That's the tough question, is how do you get insurance to pay for these things?
That's the huge, huge question, right?
In all medical devices, it's how can you fit in the existing cost structure?
So these devices now, the handmade ones that I was originally showing, they're silicone, they're pieces of wire,
they're pulled by hand.
Insurance pays about $1,000 each for them.
So the question is can we build these devices that are microfabricated, they're difficult, can we make them the same cost?
It might be a difficult question to answer, but hopefully with enough lower cost,
as flexible electronics sort of decrease in cost and decrease in difficulty,
hopefully we get close to that price point.
But right now, it's much more expensive.
What are some of the other applications for this technology that you've seen along your journey?
So definitely want to explore this area of brain-machine interface,
so looking at reliable neuroprosthetic technologies, doing cursor control, robotic arms, things like that.
We think we can have this reliable interface that isn't quite the same as spiking data,
but hopefully provides comparable utility.
We're looking at some sort of basic neuroscience questions with it.
We've done some auditory studies where we're studying large areas of auditory cortex simultaneously
while we're playing tones to animals and maybe some basic visual neuroscience systems research.
We like to work with, we're trying to incorporate optical methods into the arrays as well,
so we can, John Rogers demonstrated that he can build arrays of LEDs and photodiodes directly into the array.
So if we can build electrical and optical methods in the same device,
then we can try and explore optogenetics, stimulation, and infrared imaging,
so blood flow imaging in the same device.
Yeah, cool.
What about the use of the PCI?
Well, that's what we're going to find out, I think, is what is the spatial resolution?
Some people say you don't need it.
Some people say you need about three millimeters.
Other people say you need, probably you need maybe a millimeter,
maybe a little bit under a millimeter in human studies.
So we're going to try and basically build a device that covers a significant portion of motor cortex in primates
and see how many electrodes are useful and then see how our decoding performance degrades
as we subsample the contacts that are on the device.
So I think we won't know until, it's like spatial Nyquist, we won't know until we've oversampled.
What are your seizure models?
That's another good, yeah, good one.
So we're using very crude, very simple methods right now.
This is picrotoxin applied directly to the surface of the brain.
So it's clearly not a very good model of natural epilepsy.
And so moving forward, we're going to do chronic implants in rats
that have been treated with electrical stimulation or pilocarpine,
but still not really ideal models of epilepsy.
And so once we've sort of satisfied we can do that safely,
our ultimate goal before human testing is to implant in epileptic dogs.
So at the University of Pennsylvania we have, what's that?
I think they might even just be all population.
I think within the dog population, I think they have epilepsy at the same rate.
Some breeds might be more susceptible.
They have epilepsy at 1% of the population, just like humans.
The seizure data is indistinguishable from human EEG.
And so we'd like to implant our devices in those animals
as an ultimate test before we go into human studies.
Have you given any thought to high-density arrays in a needle format
that you could implant in the globes, pelvis, breast, et cetera?
Yeah.
I was just at the Neural Interfaces Conference,
and a lot of people are talking about very dense arrays
for deep brain stimulation.
So they're building devices now, even without this technology,
with just plain wires, very thin microfabricated wires
that have 32 or 64 contacts in a stimulating electrode for DBS,
so deep brain stimulation.
And so there's a possibility we could try and help improve that,
although when you're in a rigid substrate,
if you can make a rigid electrode, you can use conventional silicon.
The other stimulating area that you remind me of is retinal stimulation.
So there's a possibility we could make a very dense stimulator
for the surface of the retina for visual prosthetics.
Then it's a question of how many electrodes are really useful for that task.
Great.
Have you seen anything about, I guess,
biocompatibility of long-term chronic implants?
I mean, I guess the needle, some people say that it forms scar tissue
and may cause infectious disease.
And also, I guess, with your stimulation,
does it damage the tissue or cause problems over long-term use?
Right, right.
So we don't know a lot of that yet.
There is some work on passive electrodes on devices
that are just surface electrodes, just pieces of metal and wires
that have been implanted on the surface
and seem to show good biocompatibility
in that they seem to record reliable signals over a very long period
without degrading.
We don't know anything about our devices yet.
We haven't implanted them chronically yet.
We think the same materials,
as long as we can put the device together and hermetically seal everything,
which is a big challenge, we think they'll perform similarly.
The basic brain-machine interface will perform similarly.
With stimulation, I think there's a lot of safety data
from other devices about stimulation,
but we won't know,
especially when we're stimulating from hundreds of contacts.
So we're going to try and stimulate from up to 512 channels
with different spatial and temporal patterns.
We won't know about the efficacy of doing that or safety of doing that.
What can you say about kind of how was your career path to get from,
I guess you were an ECE undergrad, to a neuroengineer, a real neuroengineer?
Yeah, so I sort of randomly wandered into this, I guess.
So I did my undergraduate and master's degree in electrical engineering.
I started off in EE,
and then I worked in the wireless industry for a little while.
I went to a startup company
that was doing a new type of broadband wireless mobile data,
which was really interesting.
We had a great product, great system we were developing,
and I really enjoyed being in a startup environment.
When we were acquired by Qualcomm,
I decided to go back to grad school,
and so I went to switch fields over to bioengineering.
I thought there's not a lot in pure ECE research, I think, left,
especially in the area of wireless.
But, of course, I'm being proven wrong by NYU Wireless,
which is a new group on plugging that's at NYU.
So I changed fields of bioengineering.
I worked with Brian Litt, who's a neurologist
whose big interest is in epilepsy
and making better devices for treating epilepsy.
And so that's how I got into this field.
So I immediately tried to learn as much neuroscience as possible.
I took the neuroscience graduate courses
and tried to go from pure engineering
to learn a little bit of neuroscience,
and then continued as a postdoc in the same lab for a short time
and then interviewed and joined NYU,
where we're starting to build a little bit of neuroengineering there,
not nearly as much as you have going here,
but we're trying to grow and leverage
the strength of neuroscience and the hospital.
When you started off as an undergrad,
did you have any thought that you might end up here?
No.
I did a few bio-related projects for my master's,
but I didn't quite have this plan in mind.
It just kind of went with, I think, what happened.
But it's definitely been a very enjoyable research trajectory so far.
Okay, thank you.
Thanks.

