The Annals of Neurotech Pub: Biologists, Engineers, and Lawyers

Matt Angle:

Hi everyone, and welcome to the first episode of Neurotech Pub, brought to you by Paradromics in cooperation with SynBioBeta. In this series, I’m going to be introducing you to a number of my friends and colleagues from academia and industry, who are experts in the area of neurotechnology. We’ll share a few drinks, and we’ll talk neuroscience and neuroengineering. In each episode, I’ll choose a particular topic, and I’ll invite a few experts in that area and we’re just going to have an informal conversation. We’ll talk about some of the challenges, and some of the opportunities that we see in brain-computer interfaces, mapping the brain, using machine learning to understand the brain. We’ll just let the conversations go where they will.

I want to give you a sense for what it would be like to go out with this group of individuals for beers. I’m going to be inviting some of the people who inspire me personally. And I hope that throughout the series, some of that inspiration will come home with you as we look at the amazing things that neurotechnology has to offer, and we meet some of the people who are working to bring those technologies to the clinic and to the world.

I’m really excited about our guests today. We have Tim Harris who leads the Applied Physics and Instrumentation group at Janelia Farm Research Campus; Flip Sabes, who was one of the founding members of Neuralink; and Cindy Chestek, who is a biomedical and neural engineering professor and researcher at the University of Michigan. Today, our topic is going to be about BCI and some of the challenges behind developing next-generation BCIs. Whether it’s programmatic or technical or translational, building a next generation neural interface technology is more than just solving a single problem. And that’s what I’m really hoping that we can get into today and we have absolutely the right group to talk about it.

So I thought just to kind of warm up a little bit, one of the questions that I wanted to ask each of you was just about, in science, mentorship is so important. And I think it’d be interesting for people to hear if you had a scientific or technical mentor early in your career as a technologist, that inspired you or really kind of shaped the way that your career went. Tim, maybe we start with you.

Tim Harris:

Yeah, so I had three answers to that question. The first one is obvious, and that is my thesis advisor. So when I gave my first public talk, it got me a job at Bell Labs. So, it must’ve been really well-prepared. So, I remember very distinctly him laying out this talk saying, “You’ve got 20 minutes, you can’t go over 10 slides.” We made an outline of what was on, just a bullet list, one, two, three, four, five, six, seven, eight, nine, 10. I’m a reasonably good storyteller from a genetic point of view, but that level of organization would not have happened spontaneously.

Tim Harris:

Here at Janelia, we have a small student program and I spend a fair bit of time coaching slide decks for job talks because I don’t think my colleagues take it seriously enough. I think it is a really, really important job of how to tell a story that an audience you don’t know is likely to understand. So, I think it’s really important. But the second part of this is that, or perhaps the third, is that I had the great good fortune to go straight from graduate school to Bell Labs, when Bell Labs was the best place in the world ever at what it did. And remains at that time, the best place in the world ever at what it did. And the reason that-

Flip Sabes:

Actually, Tim-

Tim Harris:

… that was such a great mentoring experience… Yes.

Flip Sabes:

As long as you’re going to say that, there are probably going to be people listening to this, so I have no idea what that means.

Tim Harris:

Oh, well, Bell Labs was the research arm of AT&T. I’m so old that I arrived when the bell system was still intact. So I was one of 1.3 million AT&T employees. And what people know about Bell Labs is the research division. It was actually a quite big thing. It was the engineering division of the hardware equipment, but the research division was about 1200 people. It did solid-state physics and material science and high speed electronics. And it was sort of a-

Matt Angle:

And I think Claude Shannon was there‒-

Tim Harris:

Yeah, there were some important people. You know, Bode… Exactly.

Matt Angle:

Yeah. So, the transistor-

Tim Harris:

The transistor was invented down the hall.

Matt Angle:

[crosstalk] information theory, yeah.

Tim Harris:

Yeah. So, there were some important things that happened. Yeah, [Fastly 00:05:20] transform, John Tukey. So anyway, but the reason this is really important to me is that when I got there, I noticed immediately the culture. So the thing that was unique about Bell Labs as a scientific institution compared to universities, is that the maximum group size was two, the principal and an engineer. And there was an ironclad rule. You may not close your office door.

Tim Harris:

So no matter who were to ask you a question, you were sort of morally obliged to… So I got there and I have this very distinct memory of having a really great idea, had to do with the superconducting metals. I ran down the hall and asked the most senior person in my department, “Who knows about this?” And he said, “Oh, Jim Allen invented that.” I said, “Great.” Down the hall around the corner, bang on Jim’s door. He’s sitting at his desk, “Come on in. What do you have to say? Glad to see you.” “Jim, I want to do this thing. It has to do with the optical properties of superconducting metals.” And he says, “Oh, it’s really boring. They’re just ordinary metals, that optical frequencies.” And I thought, “Oh, darn. It was such a good idea.”

So, it took 20 minutes to find out who in the world had invented a field, find out my idea had no value and I could go back to work. But the point was is that that attitude of, we’re here to talk to each other, was just in the water. It was such a great thing. I feel so privileged to have been there during that period when all those people were still there, that I think, “Gosh, I wish we could reproduce that in a couple of other fields”. I would do lots of extra work to make another such place so other people… I was 26 when I got there, could experience that kind of a scientific atmosphere. Now that’s a long answer to mentoring, but it was mentoring in a different way. It was mentoring by an entire building.

Flip Sabes:

That’s great, yeah.

Matt Angle:

Flip, what about you? Did you have a formative mentoring experience?

Flip Sabes:

Yeah. So I think, although it’s trite, I’m also going to go with my graduate advisor, Michael Jordan, who is now at Berkeley. We were at MIT at the time, but he’s at Berkeley now. Mike was very much of the mind, and I completely agree with him, “You just got to learn the fundamentals. You got to do the hard work, put it in, learn the fundamentals.” Our lab meeting was three, four hours going through hard journal articles. Take the time to learn the fundamentals and everything else will be a lot easier. That was the first thing he taught us.

Flip Sabes:

The second thing was, in terms of research it was largely, “Figure it out. Do what you want.” And figure it out, do what you want, meant you had to learn to be independent, but nobody’s really independent when they start out. So what that means is, you have to learn how to rely on your peers and how to interact with your peers. Mike had a great lab, which meant that the peers that I was working with were tremendous people, many leaders in machine learning now, motor control, other areas. Well-recognized names like Daniel Wolpert, Zoubin Ghahramani, Larry [Saul 00:08:28], others. It was an amazing time in that lab. That kind of interaction of being in that kind of environment with other people, it turns out, I think in many cases, your best mentor is your peers.

Cindy Chestek:

All right. So maybe I’m going to give you three quick answers. So the first one is the person who got me into science, so that is Professor Hillel Chiel. I was a young undergraduate, EE major. I walked into his lab, I recorded from a live neuron and that guaranteed that this is what I was going to do for the entire rest of my life, right. So, that was a formative experience.

Cindy Chestek:

Obviously, my PhD advisor is Krishna Shenoy. I’m pretty sure he’s the best PhD advisor in the history of PhD advisors. We were all so happy there. It was also an amazing time in the lab. It was just surrounded by people that we all stayed in science because we were all very happy there. Like, it was such a cool environment and so there was that.

Cindy Chestek:

The other one that I’ll mention was, Reid Harrison was an excellent advisor for me because he just sort of took it upon himself to introduce me around when I was a young graduate student at conferences, things like that. I just sort of learned a lot and he had no obligation to do any of that. He was at Utah at the time, so I want to mention him as well. So, those are the three for me.

Matt Angle:

I want to use this as a launching off point. Is everyone familiar with Stevenson’s law?

Flip Sabes:

Yep.

Tim Harris:

Yes.

Cindy Chestek:

Yep.

Matt Angle:

Yeah, why did things take so long? Why was the doubling time six years? First, we can talk about a trend. But then, I’m going to ask each one of you in your own specific sort of sub areas, why things took so long. So, does anyone have a first reaction looking at this graph, to why things took so long?

Tim Harris:

I want to hear Flip’s answer.

Flip Sabes:

Yeah, okay. Well-

Tim Harris:

Because mine’s impertinent.

Flip Sabes:

I mean, I think there are a bunch of factors at play. One is that researchers… These days, I think there’s a lot more cache in… You get a lot of bang for your buck for developing technology. It’s certainly very fashioned at the moment. But in basic science, people tended to be very conservative when it comes to technology. If they know something works, they use it. So, single electrodes just worked and they worked for a long time. The first push into multi-electrode recording for many people, was something that was basically like a homemade little drive that was just multiple single-unit recordings, which you know, right?

Flip Sabes:

So, I think there’s conservatism on the one hand. And then, on the other hand, there’s just not a lot of money in making cutting-edge technology for the lab. So, it takes a lot of engineering effort. It takes a lot of concerted, coordinated effort to build things that work well at scale that push the boundaries of technology. Until recently, there just really wasn’t much of a framework for that at the level of neurophysiology, where maybe there were hundreds of labs doing this kind of work.

Flip Sabes:

These days, there are lots and lots of neurophysiology labs. There’s a lot of… People make their careers off of pushing technology. And you have institutions like Wellcome and HHMI and Allen, who decide to work together and create cutting-edge technology that really pushes the boundary. But, these things are new and I think they’re sort of of the times, but they didn’t reflect where we were years ago.

Matt Angle:

Cindy, Utah Array was invented 1988, but I think that-

Cindy Chestek:

Yeah.

Matt Angle:

… you could make an argument that it wasn’t until around the 2010s that we started to see clinically relevant and interesting work coming out of the Utah Array. You comment why did it take so long for that translational aspect to happen, even with …

Cindy Chestek:

I’m laughing because however I answer this question, I’m going to offend somebody.

Matt Angle:

I know, thats the best kind of podcast.

Cindy Chestek:

Yeah. Okay. So let me say, I think that it took a while to appreciate that brain machine interface, specifically, maybe isn’t a neuroscience discipline. It’s an engineering discipline, right? So, like the mid-2000s is when engineers actually got involved, right? So, some of these sort of early neuroscience labs, they were doing this. They did amazing stuff. I don’t want to offend anybody. None of us would be here without them. But, there’s this like treasure trove of stuff that just wasn’t applied until engineering labs sort of started taking brain machine interfaces seriously.

Cindy Chestek:

I’d also argue that neural engineering as its own discipline wasn’t a thing back then. It’s really hard to just take a bunch of neuroscientists and electrical engineers and try to get something that actually works. Now, there’s a ton of people that work right on that boundary. So, I think that that’s why it sort of accelerated recently, and it is much easier to get funded these days.

Matt Angle:

Tim, Neuropixels made in 130 nanometer process. Why did it take so long before someone did a Neuropixel?

Tim Harris:

I think three reasons. Partly Flip has identified, it is just that it wasn’t very long before I got to Janelia knowing nothing whatever about neuroscience. That neuroscience was mostly a cut-edge technology business. Nelson Sprusten, who’s at least 10 years younger than me, had to wire his own patch amplifier. That was considered a rite of passage. If you couldn’t wire a patch amplifier, you weren’t going to get a anyehere in e-physiology.

Tim Harris:

So, that was sort of the ethic of the field, is just that if you couldn’t build it yourself, you probably weren’t capable of using it well. And that slowed things down.

Matt Angle:

Jedi has to build his own lightsaber.

Tim Harris:

Yeah, right. That’s the idea. So I think that… And the second thing is just what Cindy identified, is just that it was serendipitous mixing or maybe on purpose mixing of engineers and… Well, I’m a chemist. I mean, I was trained an an inorganic chemist, which is a really broad background discipline. But, I spent nearly 20 years at Bell Labs, where the language of micro electronics was the building’s atmosphere.

Tim Harris:

So, I learned all those words, I talked to all those guys, and then I got here and ate lunch every day with the guys who were trying to record from mice. A couple of people said, “Can’t we do better?” I was just stunned at how primitive the technology was. I thought, “Man, you guys.” I mean, my first reaction, honest to God, Matt, was, “I’m going to replace 40-year-old technology with 20-year-old technology and look like a genius.”

Matt Angle:

It’s funny, Yifan Kong, who’s our CTO, he refers to that as technological arbitrage.

Tim Harris:

Exactly.

Matt Angle:

Replacing back water with sort of marginally better technology.

Tim Harris:

So that’s exactly what I did, is I wrote a spec and I wrote down what… these guys say they need this. I didn’t have any separate opinion. That’s what they needed. That’s what they said they wanted. So I wrote down, “I need a thing that does this and fits here.” And I knew enough about the subject to shop the world, to find somebody who could make it that would make it.

Tim Harris:

For Neuropixels, I think it’s just an incredibly fortunate thing that I got to be subsidized by an amazing degree because of this institution, IMEC. I mean, they have a billion dollar fab, they spend $150 million a year operating. And all I have to pay for is the hourly rates of the machines and the operators. That is way cheaper than you would normally have to do, that assemble a larger organization that could do that independently. [crosstalk] yeah, so timing and good luck is my answer to the question.

Matt Angle:

I’m really glad because a lot of your answers kind of segue into my next group of questions. One of them, this is for you, Tim, just to read off the groups that were involved in the Neuropixel project, the Howard Hughes Medical Institute, the Wellcome Trust, Gatsby Charitable Foundation, Allen Brain Institute, University College London. That’s a pretty big group of players. What special challenges come when you try to coordinate that many researchers? I know people in high energy physics and space have done that for a while, but that breaks the rules of biological research. I’m curious what you encountered when you were doing that?

Tim Harris:

So, there are three answers to that question. And one of them, I will take credit for because I am that old and that experienced. That is that biologists will make foolish suggestions about something that they would like, that no one else in the world would like and it’s impossibly hard and impossibly expensive and marginally useful. So the first thing you have to do is say, “Look, we’re just going to filter that stuff out,” because what you have to do is trust your ability to create a consensus that [Carol Swoboda 00:17:52] likes to say, “Look, you got to solve 80% of the problems for 80% of the people. The corners belong to themselves.” You can’t spend $5 million solving one researcher’s idiosyncrasies.

Tim Harris:

So, that was one really important thing. The other one was, I think just the stunning willingness of those organizations to stick their necks out. And the third thing was that it was all charities that were not trying to make any money. I mean, the contract that we originally wrote was, “And no one gets to go first and no one makes money.” It’s open to the… It turned out that we had to do a trial run and the probes stayed sort of private for a couple of years. But the written contract said that everybody in the world gets them at the same time we do. That we’re not trying to create any exclusive technology advantage through our research. It was a public spirited, five and a half million dollar bet. And I think that’s just stunningly generous.

Matt Angle:

Flip, you’ve been involved in managing a pretty interdisciplinary, kind of complex team and project. I’m curious, specifically with regard to managing the difference between doing science and doing engineering, what did you find as you were part of the early Neurolink team coming together and trying to balance the science and the engineering?

Flip Sabes:

I think, I’m going to speak more in generalities that-

Matt Angle:

Yeah, generalities is good. Yeah.

Flip Sabes:

But I think that it’s really very important when you’re taking on a big project to be clear about what the goals are. Is it a scientific goal, or is it an engineering goal? You can have multiple parts and they can be different, but it’s really important to know what you’re doing. So, I think a lot of times people build things because they can, and they’re not clear on what the goal is. Or, they’re building, engineering… a lot of academic engineering in this border where they’re just trying to push the envelope, but they don’t know what the application is.

Flip Sabes:

And, that’s fine. I mean, that’s the beauty of academics. You can do things like that. But if you’re trying to marshal a lot of resources together, either through some kind of truly unique opportunity, like the kind of thing Tim’s talking about, or if you’ve got a company, you have to be clear about what the goals are. I think that these days, if you’re trying to start under a tech company, inevitably, you’re going to have to start with a device.

Flip Sabes:

So, that’s going to mostly be an engineering effort. And so that means figuring out what the scope is. And at that point, you better know what you’re trying to do. You really should have an idea of what the applications are and what are the kinds of interfaces you’re trying to build. As I’ve said before, and I’m sure will come up again, there’s no such thing as the ideal brain interface. There are lots of different ways of interfacing with the brain. You should know what kinds of information you’re trying to get out, what scope, what brain areas, how portable, et cetera, et cetera.

Flip Sabes:

So, once you’ve sort of narrowed down what you’re building, then I think in the first instance, there’s really a big engineering push there. Now, in parallel, you could run a scientific effort where you’re sort of proving out use cases using off-the-shelf equipment, but often… And that may be a great approach for depending on what you’re targeting. One challenge with that, is if you’re really pushing the engineering boundaries, then you’re trying to build things that don’t exist right now. So, there may be no off-the-shelf option for proving out the concept.

Flip Sabes:

So, there’s this real tension, I think in general, between when do you start proving out the science or the clinical application or whatever during the engineering push? I think that is actually one of the big challenges for any company in this space. So, if for example, what you’re developing is a consumer level package for something that kind of already exists, then there is an opportunity to co-develop the science or the application and the engineering. But if what you’re doing is you’re building something that simply doesn’t exist, then that may not be possible.

Flip Sabes:

So, I think anybody who’s starting out in a company in this space, has to be aware of these issues and make a decision about when they’re going to jump in. You don’t want to wait too long to start thinking seriously about the application, or you’re just going to be building a big telescope with no idea what you’re looking at. You might find that you’ve built a telescope in the middle of a cloudy city and that’s going to be unfortunate.

Tim Harris:

So Flip, I wanted to throw in something that I had a very strong feeling for when I was first here, but less so now. And that is that I build tools for researchers. I don’t build tools for the clinic or… and they’re incredibly conservative. As you’ve said with wires, “Look, they work.” And if I’m halfway through my post doc, I ain’t changing lanes because it’s already that hard. So, when I was thinking through Neuropixels and what it did, that I had a very conscious feeling is, it needed to look like what they were comfortable with. It was, “You’ve done this before, it’s just more. It’ll be easier. But it’s not weird, it’s just more.”

Tim Harris:

I mean, there was just no barrier to adoption. As soon as they stick one end, they thought, “Okay, I like this.” But a large fraction of that was it wasn’t weird. It was the same, it was just more. It didn’t look different, I mean, the insides were really different, but it didn’t look different to them.

Flip Sabes:

Sure. Yeah, I think in some ways, electrophysiology’s electrophysiology. And if you’ve got a foreign factor of a thing that you’re sticking through the dura, then it is what it is. So, that is something that I think everybody’s going to be comfortable with. And then, you can think about, well, where do you start breaking from that and what does adoption look like? So, here’s an example. Let’s say that you’ve built a device that was so high channel count, but there was no way to get the broad‒

Flip Sabes:

Yeah. The broad wave form off of every channel. And those devices, if they don’t exist now, they will exist soon. And so, what do you do in that case? For the clinic, it’s sort of obvious what you do. You validate that it works. You show that you’ve got clinical efficacy. And the path is sort of obvious. But how does that get back to the lab? Will laboratory researchers accept it, if there’s no way to validate on a per channel basis that you are getting exactly what-

Matt Angle:

Cindy, you’ve done engineering work exactly in this area.

Cindy Chestek:

Well, and I think that this sort of goes back to the… it’s an engineering discipline. There’s a big difference between understanding the brain and building a brain-machine interface. And for a BMI, there’s an answer. Wherever you get your best information from, whatever you get a better decode from, that’s right. So, in my lab, we’ve switched completely to filtering between 300 and 1,000 Hz. Weighted average of single units always worked better than our other features. So, super low power. Dumb front end.

Cindy Chestek:

So, I think, particularly historically, a lot of these systems were super over-designed. And I have to make fun of my own starting field, which is electrical engineers, where I think they sort of heard some of these impossible demands from biologists and got really excited about it because it means they would never be done. They could get an infinite number of research grants. They could just keep plodding away at this impossible problem. And I think people didn’t sort of take a systems level good thing. And I mean, so, for a systems-level approach, you have to decide the application. And you can’t be everything to everybody.

Tim Harris:

So, Cindy, that’s a really… I mean, my colleagues here at Janelia are single-unit obsessed. And I don’t have a strong opinion as to whether or not single-unit obsession is a good or a modest or maybe an unavoidable thing. But we’re getting ready to put 20,000 channels in a freely-moving mouse, which creates a single-unit digestion problem that is really hard to cope with, you know?

Cindy Chestek:

Yeah.

Tim Harris:

So, where should I start heading? I mean, can I avoid that problem, just by avoiding that problem?

Cindy Chestek:

Yeah. Well, so, neuroscientists, for better or for worse, are fundamentally interested in neurons. They’re going to want to see the neurons. They’re going to want to sort the neurons.

Cindy Chestek:

If it’s impossible… so, we’re in a weird situation. We’re actually trying to write… my student, Sam Mason, who did this work, was trying to write a neuroscience paper. You get by far the best tuning and the best signals doing it like this because you’re getting that weighted average of the single units, you can do it after the units disappear, but we have to explain to the neuroscientists in every analysis like, “We’re doing this because now we have two thirds of the tuned channels instead of half the tuned channels,” things like that. So, if you’re stuck, I recommend doing it like that. But I think neuroscience is fundamentally different, so they’ve just got to see it.

Flip Sabes:

I think everybody would’ve said that neuroscientists, who are studying computation, are spike-obsessed. And nobody’s going to do anything. We can’t record spikes.

Flip Sabes:

But then we have these optical reporters, calcium reporters, that are too slow to record spikes, but they have all kinds of other advantages like genetic tractability and so on and targeting and lots of neurons at the same time. And suddenly, everybody’s using them. And now, of course, the technology’s caught up and you can do those and get spike level.

Matt Angle:

I think it’s important not to over-interpret some of these decoding results that say, “Okay, functionally, you’re trying to move a cursor and you don’t need spike sorting. You don’t even need spikes,” because to some extent it depends on the task complexity and the number of neurons you’re recording from. I don’t think it’s a given that for all applications, you won’t need that information.

Cindy Chestek:

Yeah, no, I mean, I think that eventually sorting multiple units might be a thing. I mean, yeah. So, at some level it has to be. In practice though… so, I mean, it’s a 10X power reduction. You want more crappy channels for BMI then a couple of perfect ones. And I really think the neuroscientists, they need the perfect ones. They need multiple sites dedicated to seeing that neuron from 360-degree angles and all that. And so, I think it’s different.

Flip Sabes:

Well, but I would even… the argument I was making was going even a little bit further, which is that… I mean, I don’t disagree with you, Matt, but going in the other direction, if I could offer you 10, 50, 100, 500,000 neurons at the kind of precision that Cindy’s talking about and you could get that across lots of areas, researchers would use it. And they’d figure out ways where the questions they were asking were well-tailored to that level of data capture. And they’d use it.

Matt Angle:

Yeah, that’s true. Yeah. I mean, Mark Schnitzer is collecting data like that. And he’s using it. You know, Cindy-

Tim Harris:

She’s frozen.

Cindy Chestek:

I’m just quiet. Yeah.

Matt Angle:

You were talking about teams and kind of programmatic challenges. And one of the things… you’ve really been pushing for translation and pragmatism. And I’m curious, starting off from an electrical engineering position and then moving to a just really pragmatic, get these things to the clinic, kind of focused research program, what have you encountered in the different ways people think about things? And how do you see those handoffs occurring?

Cindy Chestek:

Yeah, I don’t know. Well, I mean, so, first of all I think if I described my research program, I don’t think “focused” would be in there. My research program, we have stuff… our craziest stuff is the carbon fibers and slicing them in place, but it goes all the way out to our very stable, translational human work with nerves.

Matt Angle:

It’s all rooted in pragmatism though.

Cindy Chestek:

No, I don’t think it’s all rooted in pragmatism. You’re right that it’s all rooted in translation, right?

Matt Angle:

Okay.

Cindy Chestek:

This is an engineering discipline. It may look like a neuroscience lab when you walk in, but we are discovering ways of building neural interfaces. So, yeah, I mean, I think it is very pragmatic. I think it is really important that even if you’re not going to do the human research, that you have those conversations with the clinicians and just get a real sense of what you’re going to be in for and make sure you have a deep understanding of where this is all going and what has to happen. You could spend a lot of time in the lab pretending there’s a path, but there isn’t a path. And so, you can always keep an eye on that path.

Tim Harris:

So, Matt, if you don’t mind, I’ll ask Cindy a question because this is a really relevant issue to me is that the neural pixels is just about to make the transition to the technology that’s relevant to translation. I think up until now it’s been only true outliers of humans that wanted to use them for humans. They are. I mean, there are actually a number of people who have already IRB approval to stick them into people.

Tim Harris:

But I get contacted by wildly enthusiastic human surgeons. And then I ask the local environment like, “What do you think?” And they think, “Eh, I don’t think that’s really all that interesting.” And so, how do you sort?

Tim Harris:

I mean, at Janelia, I had a close community. And I could ask everybody, “What do you want? What do you want? What do you want?” and somehow sort of signal average. I’m worried that I don’t know how to signal average in the way that you’ve talked about keeping your eye on that ball of how do you understand translation in a way that’s generally rather than narrowly useful.

Cindy Chestek:

So, I would say, I think that in the neurosurgery community there’s sort of a handful of people that are really driving everything forward, and so you just sort of have to find those. There’s some neurosurgeons that are just extremely engineering-enthusiastic. And then there are some that are horrified. So, the range…

Cindy Chestek:

And actually, when I was interviewing for faculty positions, I talked to a range of people with a huge range of opinions and ended up… Michigan, the major pull was a neurosurgeon who was super enthusiastic about neural engineering. So, you just got to find those people. And they don’t exist at every institution, unfortunately.

Flip Sabes:

Actually, if I can jump on the bandwagon and ask Cindy a question too.

Tim Harris:

Yeah.

Flip Sabes:

And it’s related. At the beginning of BrainGate… was it 20… 20 some years ago, I had really mixed feelings about the move to human subjects at that point in the technical development, but since then my feelings about it have changed, but I still think that it’s a really complicated issue.

Flip Sabes:

And so, I wonder what your thinking is on, when is the right time? How do you engage? And when you’re at the university and you’re not making a product that’s going to… it’s not like you’re… even these IDEs, they’re not IDEs, these investigational device exemption approvals. They’re not IDEs to get a product to market; they’re IDEs to do science. Where’s that boundary? And what’s the difference, do you think, between when you make decisions about this at the university versus at a company?

Cindy Chestek:

Yeah. Well, so, I’ll point out, I think that a lot of first-in-human experiments happen first at a university. And particularly some of these technologies, the timeline is just so long. I mean, if it’s going to take you 10 years to get out to market, you really have to have those first human experiments happen in an academic setting. I mean, so, I like our FDA system in which it’s really on the PI to make the argument. And that argument has to be complete and put together and well supported by data.

Cindy Chestek:

And so, for in our case we had just finished the monkey experiments and on our nerve amplification surgery. And so, at that point the next clear step is to go to humans. Once you sort of… you could put together an argument that like, “I have done everything I can do. I’ve checked all the boxes.” And I think the FDA has really good expertise. The questions we got back, that’s the most expert review I have ever gotten in Washington D.C. So, I think they’re pretty good at judging these things, but it is on the investigator. And I think that’s okay.

Matt Angle:

Cindy, can you tell everyone really quickly, what is the nerve application… sorry, amplification surgery?

Cindy Chestek:

Oh yeah. So, we call it the regenerative peripheral nerve interface, but what it is, is it’s taking a small bit of muscle and attaching it to the end of an amputated nerve. And then the nerve grows into the bit of muscle. And you get what used to be a 10 microvolt signal is now a 1 millivolt signal.

Cindy Chestek:

And then we have percutaneous wires going through the skin. So, again, it’s not the device. We know it’s not the device. It’s the next scientific question. It’s, what if this doesn’t work in an amputee? So far so good in all the animal studies, but… and also, how well does it work when you’re controlling a prosthetic hand? These are all engineering questions, but you obviously can’t answer all of them in a monkey.

Matt Angle:

So, you’re using muscles as a local amplifier to pick up better signal.

Cindy Chestek:

Yes, yeah. A local bio amplifier for the nerve, and then you get nerve signals.

Tim Harris:

Awesome.

Matt Angle:

Yeah. How do we all think about peripheral versus brain implants? What do we think is going to be achievable through peripheral? When do we need a brain implant? Certainly, if you have a distal amputation, you’re probably not looking at getting a brain implant.

Cindy Chestek:

We actually did a survey on this. And it was actually prompted. There was a phone call I had with that Jack Judy when I was a first-year professor. And he said exactly the same thing, which is, “People with amputations don’t want brain implants.” And so, we did a survey. By and large, he’s right. 75% of people did not, but there’s a solid 1/4 to 1/3 of people that were actually interested.

Cindy Chestek:

So, I think that these kind of decisions are intensely personal. And so, also, in my limited clinical interactions, people are very different. And these are decisions that people have to make for themselves.

Flip Sabes:

Yeah. I think it really depends. I think there’s a technological answer, and then there’s the personal answer that Cindy’s talking about.

Flip Sabes:

On the technological side, I think there are challenges to peripheral interfaces. You’ve got soft tissues. You’ve got nerves that move around. You’ve got muscles that move around. The interfaces are not easy to build. In some ways you can think of the cortex as a breakout box for sensory nerves, for example. On the other hand-

Cindy Chestek:

Agreed. Yeah.

Flip Sabes:

On the other hand, it’s a breakout box that’s missed a whole bunch of the circuitry. So, it kind of cuts both ways.

Flip Sabes:

And so, I think that going forward and looking where technology is likely to be in the next decade or two, I think that there will be products. For example, like the things that Neuralink and Paradromics are developing. It’ll be products that would work really well for amputees. And then, of course, there’ll be products that are peripheral. So, brain implants and peripheral products that’ll work well.

Flip Sabes:

And what I am looking forward to is when patients actually have the choices that Cindy’s talking about, where you’ll be able to say, “Look, if you get the brain implant, this is easy and this is better and you can use it for other applications, but these are the extra risks. And then there are these… “ and you just look at the menu. You see what’s better. And you pick the thing that works for you.

Flip Sabes:

But I don’t think it’s obvious that peripheral interfaces are better even for a amputee. And then, of course, there’s some applications where peripheral interfaces simply aren’t going to work, like when the problem is in the brain.

Tim Harris:

Can I ask the three of you, I mean, the place where I thought that this boundary was, was in some somatosensation, is that you can clearly teach your peripheral nerves to move a hand, but it would be nice if you had a touch sensation. Is that as likely to be implemented into a peripheral nerve interface, as opposed to the paper that Richard Andersen and his colleagues at Caltech wrote two or three years ago about somatosensation interfaces that were brain implants?

Flip Sabes:

Well, Tim, there’s the DARPA HAPTIX program that focused exactly on this. And they aren’t the only people working on this, but there were a bunch of people at Case Western, at Utah, at Pitt and other places that were working exactly on this, putting cuffs… or electrodes through the peripheral nerve and stimulating. But it’s a tough thing to do. It’s hard to get access to those nerves.

Cindy Chestek:

Yeah. Well, so, I was-

Flip Sabes:

That’s what I meant by the breakout out.

Cindy Chestek:

Yeah. So, actually the project I just described was a part of DARPA HAPTIX. So, we’re on the motor side, but actually the majority of those projects were actually on the sensory side. And I think those are really cool techniques. I think it sort of follows the rules that it’s actually easier to stimulate than to record. So, a lot of those interfaces, it’s much harder to record through a cuff or a Utah array, but if you stimulate it, you actually get pretty nice sensations. So, it is something that is available, as long as you’re stimulating.

Cindy Chestek:

If you want to try recording, I think Flip’s right. I do not envy my peripheral nerve colleagues that have to deal with that structure to try to get… we try to get electrodes in, but it’s much harder than the brain. The brain is our backup plan for when we need to get spikes.

Flip Sabes:

It’s easy to stimulate, but it’s not that easy to get lots of independent channels of control listed. That’s my understanding.

Cindy Chestek:

Yeah. And that’s why we’ve basically given up on a motor for nerve. And that’s why we use the muscle taco.

Flip Sabes:

It’s funny to hear you say that it’s easier to stim than record. I understand what you’re saying in that case, but I’ve always thought that generally in our field the problem of readout is much, much easier than the problem of write in.

Matt Angle:

Yeah. I think, in my opinion, in the brain, recording is much easier than stimulation.

Cindy Chestek:

Agreed, yeah.

Tim Harris:

Oh yeah. No doubt, no doubt.

Cindy Chestek:

Though, that said, the only actual commercial applications are things like DBS, which is who cares about scar tissue, I’m just going to stim.

Matt Angle:

And DBS is the very focal form of electroconvulsive therapy.

Flip Sabes:

Don’t forget cochlear implant.

Cindy Chestek:

Yeah. Also stim. Yeah. So, who cares about scar. Yep.

Matt Angle:

Just one more thing I wanted to circle back to on the team side. I’m curious for everyone’s opinion about single labs, academic consortia, start-ups. These are the places where kind of innovations tend to happen. And do you see a clear division of labor there? Who does what best?

Cindy Chestek:

I’m curious what you guys think, but I do have an answer.

Flip Sabes:

You want to take that first, Tim?

Tim Harris:

Well, I guess that mostly the success that I’ve had in this space is entirely coordinating. I didn’t know how to do neuroscience. I just listened to people tell me, “I need this.” And my job was to go figure out how to make it become available. I don’t know how to make it either. And so, I was a conduit for, they need this and can you make that? And then it might fail. You have to be failure-tolerant. That’s okay.

Tim Harris:

But I think that from my standpoint that what I think is most important is for people to understand really clearly what their question is. You have a consortium of collaborators. And even individuals in my group, is that I’m not qualified to tell them how to do anything. My job is to make sure they understand the question that I think that needs to be answered.

Tim Harris:

And if they understand the question that needs to be answered and they haven’t told me, “That’s a really dumb question, Tim. I don’t think I want to spend my time on it,” then you just get out of the way and make sure they have everything they need to answer that question. You’re a resource conduit.

Tim Harris:

And so, I don’t see much difference in my world of boundaries between groups and across groups and across organizations, with one exception. And that is Neuropixels would not exist if we didn’t have brilliant lawyers at HHMI headquarters. No, this is not Tim hyperbole; this is a true statement. And IMEC agrees to this, is that they were tireless and fantastic. And I know of no other lawyers that would have put up with it. And because I work at HHMI, I didn’t have to pay them. I mean, it would’ve cost hundreds of thousands of dollars to get lawyers to do that if I wasn’t here. And so, there is this component of if you’re going to create an eight-member consortium with research institutions and private charities and bureaucratic English universities, it’s just a mess. And you need committed, outstanding, legal help to get to the finish line. And we had it. And so, that’s the one place where I say that you don’t have to worry about the boundaries. You really do need to get those guys on board.

Matt Angle:

The most significant inflection point in Stevenson’s Law is actually due to lawyers.

Tim Harris:

I think so. Well, so, today I found out that 5,000 Neuropixels have been shipped. And I sent my lawyer a note saying, “Chris, you’ve got to take a bow because without you we just wouldn’t be there.”

Flip Sabes:

Well, Tim, I was going to say that your observation that these boundaries aren’t so important, that maybe things have changed in the three and a half years since I retired from UCSF. But your comment about lawyers made it clear that that’s not so. I think, Tim, you’ve been in a very privileged view. You’ve had a very privileged viewpoint here.

Tim Harris:

Yeah, don’t I know it.

Flip Sabes:

And I think actually the Neuropixels story is a unique and novel model that I hope there’ll be more of. So, that’s one model. But I do think that aside from that model, there are lots of differences between academic labs, consortia and companies.

Flip Sabes:

I hate to say anything bad about DARPA. DARPA was the thing that transformed me and my lab from kind of basic science to more applied work. It funded the early sewing machine work with me and Tim Hanson and Michel Maharbiz at Berkeley. And if it weren’t for that, that project never would have started. And so, I think DARPA’s fantastic. And it really pushes the technological boundaries.

Flip Sabes:

But the big projects, I think what they really do is they get people… they juice up creativity among big groups of people, but at least in neuroscience I’m not so sure they really do what you might want them to do, which is to get lots of people working together to build a thing together. That, you can do with lawyers and with private money and with IMEC. But I don’t think that 10 academic labs can be a substitute for IMEC. And 10 academic labs can’t be a substitute for Neuralink or Paradromics either.

Flip Sabes:

And so, I think, at some point when you want to build something that’s really going to work and that you can disseminate to lots of people, either through the clinic, commercially or to labs commercially, is you need a company that can marshal extensive resources. And that’s just not what labs do. What labs do is they try out things where there’s just no clear market yet or where there just isn’t a proof of concept yet. That’s what labs are great at.

Flip Sabes:

But I do want to say, and an important thing to note, is that there’s creativity everywhere here. And so, if there are young researchers out there listening to this and they’re trying to decide which direction they want to go, there’s excitement, there’s smart people, there’s creativity, there are challenges, there’s great work across the board here, but I do think that the kinds of things that happen in academia are very different from the kinds of things that happen in industry.

Cindy Chestek:

Yeah. So, I was going to say, Flip, I agree with basically everything you just said. So, I think I’ve been in research for going on 20 years now. And I don’t think I really appreciated the difference fully between academia and industry until my research field got proper start-ups in it. So, now I get it.

Cindy Chestek:

And it’s actually, it feels really empowering in the sense that in academia we’re shiny-object chasers. And that is a super fun job. We are looking for new nuggets of knowledge. And our job is to plant a flag over it and say, “Hey, there’s a new nugget over here,” and just keep calling attention to anything new we discover. I agree, as Flip mentioned, that’s very, very different than having 40 people work on one thing and push in the same direction. Again-

Matt Angle:

I wouldn’t paint all academics that way. I mean, if I think of Thomas Stieglitz, who has just consistently pushed characterization of biomaterials in a way that maybe some people wouldn’t think is sexy, but practically to start-ups that are trying to build things that will work, that’s essential work. I mean, we care much more about those papers than we do about some sort of one-off Nature paper.

Cindy Chestek:

As do I, yeah. And I think that’s super cool stuff. Though I will point out that that is actually sort of a different lab model, to my knowledge. I mean, I think that there’s… in the US we sort of have smaller labs that chase problems individually. And so, there is a bit of a consortium there. And I don’t have any personal experience with sort of large conglomerates with lots of professors where there is kind of a leader in there. But I don’t know.

Cindy Chestek:

I mean, I think that in a lab you can have 12 people that are working on completely different visions and completely different directions. And that’s the best use of labor. If everybody’s looking in a slightly different place, you’re more likely to find something, but you’re never going to finish a product like that. It’s proof…

PART 2 OF 4 ENDS [00:48:04]

Cindy Chestek:

But you’re never going to finish a product like that. It’s proof of concept and then, you’re done and it goes somewhere else or it’s done.

Flip Sabes:

Getting back to where we started, that’s also the best way for individual investors… Young investigators to grow and learn.

Matt Angle:

I love that this is a group that’s not afraid to be controversial and I want to start with Cindy.

Matt Angle:

Where do you think people are wasting time and money, in neural engineering?

Cindy Chestek:

Oh, I don’t want to be mean, so…

Matt Angle:

You don’t have to name labs, we won’t cite to anyone, but in general, what do you think are dead ends?

Cindy Chestek:

Okay. I do have to say… So, I don’t disagree with anything Flip said. I think there’s a lot that can be head of like semi invasive signals. I’m pretty sure EEG is farmed out, right, in terms of like, real-time control. Now, I want to like specify again, what they’re good at is super controlled settings, repeated stuff, there’s so many good neuroscientists in the world. There’s still clinical applications, but real-time, single trial EEG, like I don’t think is a thing. Right? It’s hope, so not physics.

Cindy Chestek:

Was that controversial enough? Was that good?

Flip Sabes:

[crosstalk] I think the question you have to ask yourself is what are you trying to achieve? If you’re trying to achieve [PCI 00:49:17] control of the device that looks like your hand then, yeah, why would you start with the EEG? That makes no sense, at all. But, I think there hasn’t been a lot of clarity in the field sometimes about matching technology to application. So, I’m actually hard pressed. I mean, I love being a critic, it’s one of the things I’m best at and so, I’m a little surprised that I’m hard pressed to come up right now with something that I think is just clearly the wrong approach, because I really think it depends on the application.

Cindy Chestek:

Yeah. All right, I have another one. I think it has to be the end of traditional ASICs, right? Like we’re done, we have enough. Whereas, if your ASIC is an LNA followed by like an ADC and then, you want to do machine learning on your ASIC for some reason, like that’s… I think that that’s pretty traditional, at this point. It’s farmed out, right? It should be commercial or nothing.

Flip Sabes:

Okay.

Cindy Chestek:

So.

Flip Sabes:

But, you’re not saying it’s a bad approach, you’re just saying it’s not a good approach to be doing as an academic?

Cindy Chestek:

It’s certainly a bad approach to be doing as an academic. I kind of thought the Neuralink system is over-designed so, but I don’t know that much about it.

Flip Sabes:

Pretty specific.

Cindy Chestek:

Yeah. No, I don’t think you need like the front end amplifier, like I’m a big fan of… I mean like the neural dust approach, right? Like if you can do that with one transistor, just like… I think that the answer is closer to these minimalist systems that get you one crappy signal and then, you just get a lot of them. So, I think that, reconstruct it… By the time you have spent your power consumption on a whole front end amplifier, you are out of power, like you don’t have enough.

Matt Angle:

I’m not sure that neural dust makes sense, as a concept.

Cindy Chestek:

It doesn’t… Well, no, I mean… Okay, so this is the shiny object also that I get to chase as an academic, that’s like too crazy for you guys. Right? So, I think there’s a bunch of potential concepts. So, I’m not saying that ultrasound can be‒

Matt Angle:

[crosstalk] neural dust. Let’s take…OK I’ll take aim at something. Let’s talk about neural dust. Let’s talk about delivery, getting it deep. You’re either delivering it like a gene gun, ballistically and that’s crazy, or you’re sprinkling it on top in which case, what you’ve done is a really over-engineered ECOG grid. Why not just use an ECOG grid?

Cindy Chestek:

So, okay. So, let me… Yeah, yeah, yeah. So, I… I don’t know. So, this is… It’s not completely unpublished, but yeah, no, I think that you can get up and close and personal with the neurons and I think you can deliver power to thousands of [inaudible 00:51:49]. Right? So, what we’re proposing is an infrared powering scheme, something like a 200 by 200 micron chip that sits on the surface of the brain and a carbon fiber that goes deep. Right? The fun of using carbon fibers compared to some of this other stuff, is that it has a lot of penetrating power. So, I can sharpen a carbon fiber and just drive it to cortical depths. Right? So, that’s… So you can just push them in.

Matt Angle:

That’s stretching the definition of neural dust. Really, you’re talking about a penetrating micro electrode with a very small wireless, kind of ECOG.

Cindy Chestek:

Yeah, I mean, it’s got to… So, neural dust by itself doesn’t make sense. Your device needs extent so you can record. So, how… We’re getting… It’s a little like carbon fiber tail right, Yeah.

Flip Sabes:

I think, at that point, whether it’s wireless or wired, whether there’s a single LASIK or there’s a bunch of little modes, I think these are details. I don’t think they actually address the fundamental questions, to be honest.

Cindy Chestek:

What’s the question?

Flip Sabes:

What are the limits? What are the limits and where can you go?

Cindy Chestek:

The limits of… Yeah, I’m just curious.

Flip Sabes:

What I mean by that is I don’t think… I don’t… Sorry. What I meant to say was, I don’t think that these bear on where the limits are. So, you’re describing one approach to interfacing with lots of penetrating electrodes, but I think there are other approaches and it’s not clear that the dust, the wireless aspect of that is necessarily the key innovation of what you’re talking about.

Cindy Chestek:

Sure. I mean, I think people have proposed all kinds of powering systems for neural dust, right? There’s… Not all of them are impossible yet. Right? Like some of them still have runway left.

Tim Harris:

Right. Matt, I’m going to say something really, really pedestrian.

Matt Angle:

Okay.

Tim Harris:

That is, I feel physical pain for any graduate student who’s still using tetrodes because the number of hours it takes to get good tetrode recordings, I mean, they’re awesome once you get them, but you got to subtract some months to get there. So, I think we really need to do that whole field a favor and demonstrate without question that just because you can buy a spool of wire that will last you for your lifetime, it is a really bad… That is a really bad use of your time.

Matt Angle:

I spent a large fraction of my PhD sharpening tungsten wires. So, what do you think… Where do you think ECOG can go, just to give an example? where do you think we could push clinical ECOG grids?

Flip Sabes:

I think… I’m going to sort of dodge that a little bit and just say that there’s a lot of really exciting ECOG data coming out and there are lots of things you could point out, but one that’s got a lot of press and really is pretty cool is the the work out of Eddie Chang’s lab at UCSF, showing that you can get some rudimentary, to be fair, but some decoding of speech from ECOG. What you’re doing is actually decoding the speech motor signals, so it’s motor decode in the same way that sort of motor BCI is, but it’s on the speech side. I think that the precision that they’re able to get, isn’t what you would want for an everyday prosthesis, that you could sell easily in market, but I also think that what they’re getting is better than a lot of people would have thought you could get, a few years back. So, that’s just one example, but I think there are lots of… It’s a really exciting time for thinking about new ways of interfacing with the brain, the technologies that are out there.

Matt Angle:

Yeah. We’ve done some high resolution ECOG work and we were really surprised because based on some of the Buzsaki papers that we had all read, thinking about the kind of propagation of signals through cortical tissue, we didn’t expect that placing electrodes, 50 microns away from each other would yield any new information. But, I’m starting to be convinced that there’s more there than people are picking up.

Flip Sabes:

Yeah. You guys, I believe have the record now of what is it? I don’t know how many you actually record at the same time, but it was around 30K, right?

Matt Angle:

Yeah, it’s 30K. Yeah.

Flip Sabes:

Yeah. So, that’s pretty impressive.

Matt Angle:

Yeah, it’s pretty neat. It’s… I was ECOG skeptical, but when we started doing these recordings… It opened my eyes a little bit. I mean, it’s still law of diminishing returns. I would much rather have 30,000 spiking channels than 30,000 ECOG channels.

Flip Sabes:

Fair enough.

Cindy Chestek:

Yeah. Yeah. I mean, I think like… So, I agree with what you said, I think it’s surprisingly good. I also think some of the human stuff has sort of discovered this interesting zoology of units that are living on the top of the cortex.

Matt Angle:

Do you want to comment on that? Well, what do we think that is? Do we think those are axons do you think they’re like…

Cindy Chestek:

Yeah, that’s what I was going to… I think they’re layer one neurons, right? There’s enough neurons on there, you’re not necessarily getting a spike on every channel. So, an undergrad in my lab actually did a model of this and there’s just no way you’re getting a somatic, like a somatic spike from a millimeter away. Yeah, you need new physics.

Matt Angle:

It’s not clear if there’s some weird ectopic layer one cell that’s hanging out and you can record a spike from whether that’s decoded or not.

Cindy Chestek:

But, there’s neurons up there. We sort of… We did our own histology for that, just to check and it’s like Christmas lights, there’s neurons up there. Right? The epilepsy people study those neurons, all the time. So, I think it’s probably just layer one neurons. Right? What that would also mean is that you got to stay close. Right? So, if you start… If you develop a layer of scar tissue, I’m curious if you have a solution for that, because once you’re 50 microns away, that’s going to be a problem. Right? Do you have an answer?

Cindy Chestek:

I’m curious, is there anything new or…? Because it’s going to develop a scar, yeah.

Matt Angle:

Yeah. No, no. That’s… I mean, from a clinical standpoint, that’s not where we’re going. We’re not building high-density ECOG grids. We had an opportunity to leverage the Argo to collect some of that data, but we’re going for penetrating microelectronics.

Cindy Chestek:

Penetrating? Cool. Yeah. 1/r is just really hard to get around, so.

Flip Sabes:

That’s true.

Tim Harris:

Or R squared or R cubed.

Cindy Chestek:

Usually, with neurons, it’s one over R, I would argue because the R squared, you have to be like a millimeter away before it starts really being a dipole.

Matt Angle:

What… Let’s talk… Let’s talk about other modalities. What do we think about optical? What do we think the time horizon is before optical interfaces replace electrical ones, as that go-to for, let’s say clinical, because in some ways, I would say that in research they’re both there and they’re both producing great data? But, when do we think they’d be viable for clinical use?

Tim Harris:

Matt, for reading or writing or both or neither?

Matt Angle:

Whichever you want to answer.

Tim Harris:

For clinical‒

Cindy Chestek:

Yeah, I’m a bad person to answer this because I have very little biological knowledge, but I mean, you’re talking about genetically modifying human beings, right? Which I mean, not saying that that’s necessarily a negative, if you’re talking about a very serious illness, that’s obviously already being done, but yeah, I’m not a good person to answer this.

Flip Sabes:

I think it’s always a risk benefit, trade off. The big advantage of… It’s not really so much about optical per se, but about transgenic approaches, is that you can get cell type molecular specificity and the clinical benefits of that are potentially huge. So, there are clinical trials now, certainly in the retina, and I don’t… Off the top of my head, I don’t have a full list of them, but there are people who are working in this domain.

Flip Sabes:

I think we will see those technologies continue to advance. It’s tough because the potential risk points are things that could happen quite a bit down the line. So, it’ll be decades before we can say, with surety, that the kinds of things that people are doing now are safe, when you talk about potential for, say, development of cancers or something. That said, if the use cases, it makes sense that you start out with people for whom that’s an acceptable risk, because there’s either a severe quality of life or a life and death need now. So, I think it won’t be that long and what does that mean? Does that mean five years? Does that mean 10 years?

Matt Angle:

I guess the question is, let’s imagine, for example, a patient with tetraplegia due to ALS or a spinal cord injury, that’s sufficiently serious that the function will never return. So, the area of the brain is not connected to anything. I think you could make the same benefit risk analysis either for a penetrating microelectrode or for a gene therapy approach. Do we think that there’s a chance that some optical approach will come leapfrog Neurolink and Paradromics and an image from a million neurons and a square centimeter of cortex, is that…?

Flip Sabes:

Well, Tim, let me ask you a question about that, you’re really the expert here. It seems to me that there’s no free lunch here. I mean, there’s still the energy requirements either to pump out a lot of light or to run the ASICSs required for cameras, basically are still high. So, you’re still going to run up against heat dissipation limits and so, it’s not as if, oh, if you could only stick these into a human, now we could be recording from a million channels in human subjects. It feels to me like there’s still some kind of fundamental power limits. What’s your take on that?

Matt Angle:

Yeah, the two photon approaches are very power consuming and the scanning is necessarily slow. It’s usually like few scan points, so you’re not getting great time resolution. I think Conrad [Cording 01:01:31] and Adam [Marblestone 01:01:32] did some kind of physical modeling. I think it depends on your requirements, like if you… Maybe, if you say, I want to do widefield calcium imaging, I don’t care as much about the timing and I take a hit on the signal to noise a little bit, to lower the power.

Flip Sabes:

So, optical ECOG.

Cindy Chestek:

And we’re back to ECOG.

Matt Angle:

Yeah, you’re right. I don’t think there’s a free lunch. I think there’s some interesting… Vincent [Parabon 01:02:06] is doing some really interesting bioluminescence work right now. I think if you can… If the signal is sparse and originates locally, you solve a lot of your problems, but I don’t think anyone’s solved it yet.

Tim Harris:

Well, I think that Matt, you… If the problem is sparse, then there’s a reason to go there because penetrating electrodes don’t do sparse very gracefully.

Flip Sabes:

Right.

Tim Harris:

Optics do sparce much more gracefully, but in addition to all of the sort of power requirements is what I’ve always said is if brains were transparent, I’d be building microscopes, but they’re not. So…

Matt Angle:

Yeah.

Tim Harris:

We’re doing what we can, even though… If you’ve come from outside the space like I did, and you ask people, how are you going to do this? The first answer is, well, we’re going to poke a hole, that doesn’t strike you as great start. But, on the other hand, it seems to be okay and the alternatives are less appealing. So, you poke a hole.

Tim Harris:

I mean, you do what Cindy does and try to figure out how to poke the smallest possible hole.

Cindy Chestek:

Smallest possible hole.

Tim Harris:

That certainly progress over one channel and a really little hole is clearly progress over one channel and a pretty big hole, like a Utah array makes. But, still, my notion of displaced volume versus channel count, I think high density silicone probes, just win, hands down, that issue of how much tissue are you going to displace. Now, there is a size issue was that little seems to be benign at some dimension that is of the order of carbon fibers. So, in terms of very long-term tolerance that kind of dementia may end up winning.

Cindy Chestek:

Yeah, no, I mean, I totally agree that, I mean, I think silicone and I mean, it totally wins every time on channel count and I mean, any… I mean, and Matt, your system, like it’s interface to a silicone backend, that’s just going to be much more higher channel gap. I think where the ultra micro electrodes and carbon fibers are just one of them come in, is say I want all of it, like I want centimeters and centimeters of brain. I want thousand channels distributed, and I want it to be perfectly safe, right? That’s sort of where… That’s going to be a long time coming, but that’s the goal. [crosstalk] I think carbon has a chance, yeah.

Matt Angle:

I think, the other place where ultra micro electrodes do very well is the kinds of coatings you can use for something that’s inorganic and tolerant of high temperature.

Cindy Chestek:

Sure.

Matt Angle:

If you have active silicone, your ability to protect that in a really nasty biological environment is still… I haven’t seen that technology yet. I know IMEC’s working on some cool nano laminate stuff.

Cindy Chestek:

Yeah?

Matt Angle:

Flexible polymer based electrodes. Stuart Cogan’s been looking at how to coat that in silicone carbide. Michel Maharbiz has been looking at silicone carbide for a long time. Elon tells us that he’s looking at silicone carbide now, so that problem’s solved. But, in reality, that material challenge is kind of still unsolved. I would say if they… If the thin film issue, if you can have a thin film that has a real… has really low water vapor penetration, can last for 10 years, I’m on board. I would take a neuro pixel, I’d wafer thin it and I’d make 10,000… I put 10,000 neuro pixels in, but I think until that’s a solved problem, you’re going to be looking at PtIr carbon fibers.

Cindy Chestek:

Yeah, no. If I was going to write a science fiction series where like suddenly one new technology turned up and changed everything, it would be the magic coding.

Matt Angle:

Yeah, yeah.

Cindy Chestek:

Right? That suddenly lets you make… All of your medical devices can now be chip scale, right? Because you have the magic coating and so, it’d be great if we had it. I wonder if material science has been as fully engaged as they could be, maybe it’s out there somewhere. I don’t know.

Matt Angle:

Well this is a great.

Flip Sabes:

Sorry, go ahead.

Matt Angle:

I was just going to say, this is a great point.

Flip Sabes:

I was just going to say earlier though, that in response to Cindy, that obviously the coatings and longevity are an issue, but I don’t think it’s the only really hard technical problem. Delivery remains a hard problem. If you’re going to be putting that many devices into the brain, you still got to worry a little bit about blood-brain barrier. You still got to worry about wires. You’ve still got to worry about processing and heat dissipation. So, I think there’s… The good news is that there are a lot of really interesting problems for the engineers who are going to be watching this. It’s not just the [crosstalk] people.

Matt Angle:

Although, for the scales we’re looking at, I mean, we’re comfortable with where the heat dissipation is right now. I mean, partially due to Cindy, also Takashi Kozai work and you share common‒ Ross Patel. All of the work with carbon fibers I think is really encouraging about the blood-brain barrier. If you get small enough, that data looks really good. Cindy, I don’t know if you want to.

Cindy Chestek:

Yeah, well, so I was going to say, I mean, I think that… I think there is a chance that you can get small enough and maintain enough penetrating capability that you can get in there without breaking blood vessels. Right? I’m also basing that on Nick Melosh’s lab at Stanford, also. I’m not sure. I think the delivery problem is much, much harder for soft devices. I’m not saying, like the devices may ultimately need to be soft for this to be completely, utterly biocompatible. I don’t know, but maybe not, maybe you can get, whether it’s carbon or tungsten or something hard, platinum iridium, you can actually get pretty far below cellular dimensions and still get something into the brain. Right? So…

Flip Sabes:

I agree. I totally agree. I was thinking about scaling up because you were talking about the science fiction. So‒

Cindy Chestek:

But, why can’t I do that 10,000 times, right? If the first one doesn’t break a blood vessel and I put them far enough apart, right? I mean, say I put them 200 microns apart and if you’re not breaking capillaries, you’re not breaking capillaries. Right? So.

Tim Harris:

Cindy, I learned a really hard lesson about that when I was in venture and it goes back to Flips notion of I’m a complete outlier when it comes to the environments that I lived in because I did Bell labs and then I did, generously funded venture and now, I’m at HHMI. I am now a part-time professor at Hopkins. So, I am getting a taste of that space, but the reason I brought this up was that I used to think if you bought portable hard drives that they worked and then, I bought… Then, I started buying them at the rate of a hundred a month [crosstalk] If you bought a hundred portable hard drives a month, you find out that many of them don’t. So, the fact that you can put a few carbon electrodes without breaking a blood vessel is not convincing to me because I want to see many hundred [crosstalk] many time.

Cindy Chestek:

Yeah. Isn’t that being done thought, Matt?

Matt Angle:

Yeah, maybe the counterpoint is also take the Utah Array. The Utah Array is, when it implants, pretty messy, to be honest. It causes a lot of disruption, but it’s allowing people to move cursors, move robotic arms, decode handwriting, and it lasts for years. So, I mean, I think it’s important not to get academic about this, you don’t want to take the position that you can’t disrupt the blood brain barrier, or you don’t have a product. I think you need to understand… Yeah.

Tim Harris:

Yeah, good enough is good enough.

Matt Angle:

Yeah.

Cindy Chestek:

Yeah.

Matt Angle:

And better… Right now, better than the Utah Array means good enough.

Cindy Chestek:

Yeah. Another topic is you were asking earlier about how… When is it the point to leave the university? One of the things that I think would be a big disruption would be for universities to suddenly have access to implantable packaging, right? If we had a general purpose medical device, if we didn’t have to go to Medtronic and go through the lawyers to try to use something for a different application, that is something that could theoretically happen in the next 10 years that would change a lot and would enable a lot of first in human experiments.

Speaker 1:

No, packaging is hard.

Flip Sabes:

Yeah.

Matt Angle:

Yeah.

Cindy Chestek:

Yeah, no, packaging is very hard, but I mean, even conventional packaging becoming available, the Utah Array on a package, like whatever that device is, if it was broadly available, if there was even a hundred channel system you could buy, that would really change everything, I think,

Matt Angle:

Yeah. Utah Array, I mean, speaking from experience, building a similar to Utah Array in a hermetic package with features that will last 10 years, that’s an integration challenge.

Cindy Chestek:

Mm-hmm (affirmative).

Matt Angle:

It’s absolutely doable, but it’s not the kind of thing that a single lab will be able to do. Yeah, a hermetic feed throughs is a thing. I would love to see… I would love to see a generation of better feed throughs because for form factor devices like ours, that would mean a generation of better brain computer interfaces.

Cindy Chestek:

Yeah, I’m hoping they do at some point decide that packaging is DARPA hard and have like DARPA challenge, $75 million for packages and everybody would be super into it.

Matt Angle:

Let’s talk a little bit about application. Where do… Where do you see the ball going, in terms of application? Is there something that you see that is accessible right now from a scientific point and is only gated by engineering? Not like an obvious thing, like repeating BrainGate, but I’m curious if you think there’s something that might not.

Matt Angle:

BrainGate. But I’m curious if you think there’s something that might not be obvious to the listeners, that you think is very scientifically achievable, and the only reason that doesn’t exist is because no one’s done the engineering.

Flip Sabes:

I actually think there are a lot of applications where they’re really strong suggestions, that’s true, but because the engineering isn’t there, the science isn’t there, the complete story of the science isn’t there. A really good example is DBS for depression. DBS for depression works. There are clinicians that use existing devices off label fairly regularly. There are lots of patients who’ve benefited from it. The clinical trials haven’t gone that well. And part of that, I think that there are lots of reasons.

Flip Sabes:

There’s actually a really interesting story in the Atlantic if you’re interested in it about that. But one of the, there’s some non-scientific reasons, but I think there’s some scientific reasons as well, which is that the variability of the response is not as good as you might want. And so here’s an example where if you had the ability to target more sites, and if you had good closed loop recording, then I think it seems really clear that you would have a better device, you’d be able to get better and your success rate would go way up. And not only that, the indicated population would go up. But I wouldn’t say that like the science is a slam dunk. It’s obvious it’s there. It’s just that the data will look really good for it.

Cindy Chestek:

Just trying to think of one. I think maybe visual prostheses are sort of more on the sort of engineering challenge at this point, like there’s enough real estate in the brain. There probably are neuroscience questions, we’re not quite sure how to stimulate in the best possible way, but that one strikes me as, if we deployed tools similar to the ones we have with sort of enough of a push behind it, then that’s something. I watched that news and I expect more to be coming out soon. Cause I think it’s very doable.

Flip Sabes:

Although even there, Cindy, it’s a good example because nobody knows what it would look like to have millions and millions of kind of phosphenes, right? Does pixelated vision look good or? I mean, certainly you’re going to be able to see more than with nothing, but how good does it look? And then there are people who think about also putting electrodes up and higher visual cortex and using the hierarchy and filling in. And so that’s just another good example where I think there’s a clear win and if we could put it in a, if I couldn’t see and you could put a million electrodes into my V1 I would do it like that. But I think there’s also a lot of really interesting science and a lot of opportunities.

Tim Harris:

Yeah, Matt. I make tools for rats. I don’t have qualifications to speak in this lot.

Matt Angle:

Yeah. We’re going to listen to Frank. This is Frank Willett from Krishna Shenoy’s lab. He has done some pretty amazing work in most recently decoding imagined handwriting.

Frank Willet:

So I’m just kind of curious what, I don’t know if this is a good question, but I guess I’m just curious, what is the fundamental constraint on channel counts and why can’t we have hundreds of thousands and you know what…

Matt Angle:

So what are the physics here? What are we up against

Tim Harris:

Power?

Cindy Chestek:

That’s a major one. I was thinking about that one. Yeah.

Tim Harris:

Heat.

Cindy Chestek:

Yeah.

Tim Harris:

And how many holes are you willing to poke?

Cindy Chestek:

Yeah. Well, yeah. I mean, I think again, I feel like I know a way out of the holes more than I know a way out of the power, right? Like that’s the power consumption is legitimately challenging and that’s why I think that we just have to consider much different looking architectures for the front end, if we ever want to see channel counts, anything like that.

Matt Angle:

What is your per channel power right now? Cindy?

Cindy Chestek:

Oh sorry, I don’t have a good number off the top of my head. It’s about a 10 X reduction from like from a 30 kilo sample system. Right. So, but it’s still not enough. We can only get our chip down to about 200 by 200 microns. Right. So it’s about, okay, so one channel one microwatt optically delivered.

Flip Sabes:

Power is also an issue though on the backend, if you’ve got enough data and at some point there’s limit, you’re not going to be able to get stuff off the head, even if you’re just sending timestamps. And so then that means you have to do the relevant competition on the head and that poses a challenge because you’ve got more and more units, which means the models will become more sophisticated. There’s more front-end processing. And so there’s a little bit of a non-virtuous, some word for that, cycle.

Cindy Chestek:

Vicious.

Flip Sabes:

And there you go, vicious. By proxy again. Yeah, there’s a vicious cycle there. So I think power is clearly a big one and tissue displacement. I mean, yeah, we can. It’s true. Of course, that you can make these things smaller, but Cindy, your devices are only single channel per thread. At some point, even those small threads start to display dislodged tissue.

Matt Angle:

Tim, have you started looking at what it would take to put many, many, many neuro pixels in a critical area?

Tim Harris:

Well, I’ve started thinking about it and worrying about it. And I ended up right where we’re flipped, just left is that I think we can make very small devices with very high capacity, but in a research environment, I worry about, can you get them in? I mean, they’re getting so small that just the whole notion I’m going to pick it up and I’m going to glue it to a stick and I’m going to put it on a microdrive.

Tim Harris:

And I mean, none of that works when you get to the kind of capacities that Flip has been worrying about. And so the question is, how do you create an engineering environment where you can, let’s say you’re going to put, I don’t know, 500 shanks into a rodent, or even a primate. Everybody wants their 500 shanks in a different place. And so you’ve got to decide we’re going to put 20 in this brain region and 50 in that brain region. And so you have to create an environment in which the user can cope with that challenge. And it’s not a hundred million dollar investment. And so I’m struggling with that right now and we’re going to do it [crosstalk]

Matt Angle:

Flip, that’s precisely what you worked on. Right? Can you tell us about the sewing machine and then we’ll pick at it?

Flip Sabes:

Yeah. So the whole idea of the sewing machine, now keep in mind that when we started thinking about the sewing machine, I was very much thinking about primate and then human. We were not thinking about rodent. It’s a very different thing. And the question was that the technologies that were out there, Utah array and others as well, were limited to sort of fixed arrays or small number of electrodes. They couldn’t go very deep and you couldn’t get this kind of like custom mix of density and breadth, which I think is what you probably want for the kinds of problems that I was looking at as a scientist at the time it actually, originally the motivation for working on this was we had scientific questions. We wanted to ask about how, for example, parietal cortex, frontal cortex, and other brain areas work together in sensory motor control.

Flip Sabes:

But then it quickly became clear that if you could solve those problems, you’d have a great PCI as well. And so what we’re thinking about is how do you deliver individually targeted electrodes or smaller rays of electrodes to say hundreds of sites in tens of brain areas. And that was the set of criteria that we started with. And I was really lucky at the time because Tim Hanson, I’d just gotten some DARPA money to work on problems in the space. And Tim Hanson had been an engineer working with Miguel Nicolelis at Duke, and he was available and came on over, but I wanted him to like do some programming and stuff, so I’ll come over, but I want to work on this problem of delivering lots of really fine electrons.

Flip Sabes:

And at that point we brought Michelle on and three of us sort of brainstormed for a long time about what were the right approaches. And we sort of came up with this idea of thin-film, individually delivered robotically, and that solved lots of problems. The thin film, by the way, just to be clear that the thing that it solves is because it’s a thin film, it’s easier to integrate. You can make lots of the same time. It’s easier to say, for example, on a tip to the backend. I mean, the packaging problem remains hard, but it’s a little bit easier to say than carbon fiber. Yeah.

Cindy Chestek:

Well, so let me actually say, let me not pick at your system. Let me use your system to push back on one of your previous points. I don’t think you have a volume displacement problem. When you’re talking about a nonhuman primate brain and you’re talking about let’s say thousands rather than tens of thousand channels. I mean, I think it fits, right? If you make it really small, you’re not displacing that much tissue. I mean, you can get a lot of electrodes in there before you’re up to one Utah, right?

Flip Sabes:

Yeah. There’s no doubt that if you could magically wave a wand and get the devices that you’re talking about into the brain, displacement would certainly not be limiting factor. All these other issues about packaging and cabling is going to be much. I actually agree with that, yeah. But I guess what I meant by that, I was also thinking about the delivery problem.

Cindy Chestek:

Well, yeah. I mean, I guess, yeah. So the fun of the carbon fiber is you don’t have a delivery problem. Right? You can just push it into the brain, right. You can’t get to like deep depth. So it is a critical solution at that point, when you’re talking about carbon fiber alone. But it does just go in.

Matt Angle:

Yeah. If you want to insert 10,000 of them at the same time, you do have to be a little more clever.

Flip Sabes:

Yeah.

Cindy Chestek:

This is true. But by the time they’re 200 microns apart, they’re fairly independent, right? Like we’ve never had an issue with that.

Matt Angle:

Have you ever gone to 10,000?

Cindy Chestek:

No, but…

Matt Angle:

Because we have.

Cindy Chestek:

You had breakage, right? Like I’m actually really curious. What…

Matt Angle:

If you want to insert 10,000 ultra micro electrodes in parallel.

Cindy Chestek:

Yeah.

Matt Angle:

You have to, you have to ablate the Pia.

Cindy Chestek:

Okay.

Matt Angle:

In our hands, you can’t do it with an intact Pia. It’s possible. And we developed a laser ablation system.

Cindy Chestek:

Sure. I mean, I don’t want to get like super technical here, but weren’t you closer to 20 microns in size and didn’t you have glass, right? Wasn’t there breakable components?

Matt Angle:

Not glass. Melosh lab uses glass. We…

Cindy Chestek:

So it’s pure platinum Iridium and it’s below 10 microns.

Matt Angle:

We’re sub 20 but more than 10, so I’ve never done 10 micron 10,000, but I don’t suspect that that’s going to be fundamentally different. But I think that we thought, you put them far enough apart from one another and they insert individually. We thought that that was a sort of…

Cindy Chestek:

I mean at some level that’s true. I mean, you might not be far enough apart at your size and this might be… I mean, its the Melosh lab that has sort of made this a quantitative science.

Matt Angle:

I’ll be very excited to see the 10 micron probes, 10,200 microns apart in cert without manipulation of the Pia, but I’d be willing to make a little scientific bet that that won’t be the case.

Cindy Chestek:

Well, so I think he just gave me another shiny object to chase.

Matt Angle:

Absolutely.

Cindy Chestek:

Yeah. Fun. Yeah. No sharpen carbon fibers are particularly cool. My student Alyssa Wellie has been sharpening them and it’s so much easier to insert in everything. You know, it’s very similar to Tim Gardeners, but now we’re doing it for sort of individuated probes. Like arrays.

Matt Angle:

Yeah. And that’s something that came out of Nick’s sort of mechanical…

Cindy Chestek:

Sharp matters.

Matt Angle:

Sharp is important.

Flip Sabes:

The recent papers, Abdullah Bates paper is really great. Excellent.

Matt Angle:

Abdul is awesome. Is there anything you think we should’ve talked about today that I didn’t ask. Do you want to ask each other any questions? Oh yeah. I can play circus question, but we sort of got to it already with the magic coding. But if you want to play it, play it. We will give him his airtime.

Sergey Stavisky:

So all three of these experts at your next show, they all work on the hardware interfaces as do you. Of course that. And a question I would want to hear more about is, so a big question with all of these devices, device life span, how long will they last in the brain? And so I’m curious, what evidence would tell you that a design for neural probe is going to last a decade or more. Of course I want extreme, there’s you put it in and you wait 20 years and if it still works, it still works. That’s maybe not the most pragmatic answer. So, before that, what would you like to see to have high confidence that a device is going to be suitable for long-term use?

Tim Harris:

I already said, I build tools for rats. They don’t leave 10 years. So I’m not involved.

Cindy Chestek:

So, this is another, I think it’s another example of “I’m glad neural engineering as a field. And now people are actually trying to like figure out how to build better soak tests” and things like that. I mean, I think a hot acidic soak. You can debate like how extreme that should be. But I mean, by the time you have survived a really aggressive soak test. That’s a really good indication. So you have to not just have some channels survive it, it has to be like your whole system survives it. But yeah, I believe soak tests.

Matt Angle:

There is a really exciting accelerated reactive aging project going on at the FDA. And that shows promise, but we shouldn’t delude ourselves to think that there is like a… There’s not like a one-year test that will tell you if you’ll last 10 years in vivo. [crosstalk] there’s like a lot of exciting research, but it doesn’t…

Cindy Chestek:

Yeah. There’s a lot of things that you should, the bars you should probably get over. Right? And then…

Flip Sabes:

So, as scientists, we know you can’t validate a test without validating the test. And so if you want to know whether a test is good to tell you whether something last 10 years, it’s going to take a decade to know the answer to that.

Matt Angle:

That’s right.

Flip Sabes:

And so, but you’re right, Matt, that the work going on with the FDA is great. There are other people developing these tests. I think probably my guess is we’ll get to, is it’s going to take time for the reason I just said, but we’ll have a bunch of different sort of validated tests. Some of them will be soak tests and maybe mechanical tests the people like, and then there also be In Vivo marker. So for example, the problem isn’t just about the chemical environment. There is also the tissue mechanical interface. And so at some point we might decide that let’s say, for example, if you could put a device in just throwing out a number there, I’m not advocating on any ethical grounds, but let’s say if you could put a device in 10 pigs and it lasts a year, then that’s a really good indicator that it’s going to last 20 years in a human or something like that.

Flip Sabes:

And the failure rate on those 10 gives you some idea about what’s going to happen in time on a human. So I think this is an area that is extremely un-glorious and there’s not a lot of, and it’s long timescale. So it’s really hard for academics to do that. So it’s an interesting question is how does the community work together to develop these standards?

Matt Angle:

Yeah. I mean, it’s also really hard for startups. I don’t know any investor that I couldn’t pitch on ten year soak test.

Flip Sabes:

Right, right. That’s interesting. Yeah. Something to think about. The FDA is a great place to do it, obviously

Matt Angle:

I’d like to see Janelia, honest to God. I think Janelia would be the place. Look at what they’ve done with calcium indicators. If Jerry Rubin said I’m going to make the ten-year magic coding, I think they could do it. I think if they work with Imec. I bet they could do it.

Tim Harris:

I’ll talk to Ron Vale and see if he’ll buy it.

Flip Sabes:

Well, actually, that’s a great lead into the question I was going to have for Tim.

Tim Harris:

Okay. So ask away.

Flip Sabes:

We’ve talked a bunch about academic environment, talked a bunch about industrial environment, but we’ve always talked a little bit about that kind of weird, special perch that you had there in developing the neuro pixel at Janelia. So what’s next? Are there other similar, I mean, obviously you’re going to iterate and improve on the nerve itself. But are there other similar kinds of projects that people have talked about bringing together public, private or all private money to solve big technical problems that are used widely?

Tim Harris:

No, I mean, we had a competition that in the end didn’t select a winner, but the HHMI said, “we will invest 250 to $300 million over 15 years if you give us a good enough project”. And it was a straight up challenge. We’ll write the check, you have to come here and do it. And so I think that idea is still on the table, it’s that the foundation is not stuck where we are. Their notion is we built this place so that it wouldn’t become a standalone stable institution. We wanted it to change continuously as to what it was trying to innovate.

Tim Harris:

And so I think that I encourage you all to think hard on those. What could, what could $250 million in 15 years do for the whole community at large, that just is unlikely to happen anywhere else? That my worry is that when I first got to Janelia, somebody from DARPA came to visit me every three months for two years. And we just never sink because I told them “when I try something, I expect it to work”. And as far as I can tell, you choose projects, you expect to not work, and there’s not much overlap.

Matt Angle:

As someone who received $14 million from DARPA, I’d like to refute that. I’m actually obligated to refute that.

Tim Harris:

No, but I mean, there is this attitude that we’re going to shoot for Mars and hope we hit the moon. And I think from an engineering standpoint, it’s really hard to succeed if you don’t hit what you aim at. Once in awhile you hit Mars, but mostly you don’t, and mostly you don’t hit the moon either. And so, no, it’s just the personal preference is that if I’m going to work 75, 80 hours a week for a decade, I want to be pretty sure that I’m actually aiming at a hittable target.

Matt Angle:

Yeah, I mean that’s one of the challenges though, of applied science versus engineering.

Tim Harris:

Well, you know, I applaud your thought and the answer is the question is on the table. What shall we do with a big budget and 15 years to do it?

Cindy Chestek:

I don’t know. I mean, I actually think that the moonshot really has a place in all this. I never have more fun than when DARPA asks for something impossible, right? Like the week that follows DARPA asking for something impossible, even if we don’t even end up applying to the impossible thing. You just sort of get in this mindset of like, “okay, what if, what if” and so I think they form a really important role. But I agree. It doesn’t actually end up accomplishing the thing.

Matt Angle:

Yeah. You know, actually DARPA’s annoying insistence on things that weren’t possible was to some extent drove our ASIC design. I think one of the things that we did, that’s quite clever is developing a sort of analog front end for doing principle component analysis. Just simply because they demanded spike, sortable data on all channels, and they demanded a shitload of channels. And we told them, “no, you can’t do it”. But actually the result of their insistence was that we went back to the kind of first principles and realized that some of what we were asserting was kind of dogmatic rather than first principles. Now, at the same time, they asked for a lot of other things that weren’t possible that weren’t possible, but sometimes that rooted insistence and trying things that are crazy is useful. Well, thank you all.

Cheers.