On this episode of the AI For All Podcast, Nandan Nayampally, CMO at BrainChip, joins Ryan Chacon and Neil Sahota to discuss edge AI, neuromorphic computing, and AI hardware. They talk about the benefits and trade-offs of edge AI, neuromorphic chips and computing, the economics of hardware, top AI applications, AI in health, and future opportunities in AI.

About Nandan Nayampally

Nandan Nayampally is an entrepreneurial executive with more than 25 years of success in building or growing technology businesses with industry-wide impact. He was at Arm for more than 15 years in a variety of product marketing and product management leadership roles, eventually becoming vice president and general manager of Arm's signature CPU group and the Client Line of Business where he identified key technology investments, developed a strategy and roadmap of products to deliver compelling, market-leading solutions for billions of SoCs, while establishing strategic partnerships and alliances. Nayampally comes to BrainChip from Amazon, where he helped accelerate the adoption of Alexa Voice and other multimodal services into third-party devices.

Interested in connecting with Nandan? Reach out on LinkedIn!

About BrainChip

BrainChip is the worldwide leader in edge AI on-chip processing and learning. The company’s first-to-market neuromorphic processor, AkidaTM, mimics the human brain to analyze only essential sensor inputs at the point of acquisition, processing data with unparalleled efficiency, precision, and economy of energy. Keeping machine learning local to the chip, independent of the cloud, also dramatically reduces latency while improving privacy and data security. In enabling effective edge compute to be universally deployable across real world applications such as connected cars, consumer electronics, and industrial IoT, BrainChip is proving that on-chip AI, close to the sensor, is the future, for its customers’ products, as well as the planet.

Key Questions and Topics from This Episode:

(01:13) Introduction to Nandan Nayampally and BrainChip
(01:25) What is edge AI?
(04:09) Benefits of edge AI
(06:25) Edge AI trade-offs
(08:23) Is hardware the next frontier?
(10:38) Investment in neuromorphic chips
(13:31) What is neuromorphic computing?
(15:32) Impact and economics of hardware
(18:48) Long-term benefits for business
(21:12) Implications of edge AI and decentralization
(24:58) Top AI applications
(26:16) AI in health
(30:28) Data sources for enterprise AI
(34:16) Future opportunities in AI
(36:28) Learn more about BrainChip

Transcript

- [Ryan] Welcome everyone to the AI For All Podcast. I'm Ryan Chacon. With me today is Neil Sahota, my co-host. He is the AI Advisor to the UN and the founder and so-called Godfather of AI for Good.

- [Neil] Hey, how's everyone doing?

- [Ryan] I also have our producer, Nikolai, with us as well, who's gonna be chiming in with some pretty good questions, I'm sure, throughout the conversation.

- [Nikolai] Hello.

- [Ryan] Today's episode is all about AI chips, AI at the edge. Lots of good conversations here, and we all know companies like NVIDIA have made huge strides in scaling up AI power, but what happens if you want to scale down? And that's one of the questions we're gonna talk about today. And to just discuss that, we have the CMO of BrainChip, Nandan, here with me.

They are a global leader in edge AI on chip processing and learning. Nandan, thanks for being here.

- [Nandan] Hey, thanks Ryan. Thanks for having me on.

- [Ryan] Yeah so let's start off talking about AI at the edge and this idea of scaling down AI. What does that look like compared to AI off the edge and what does that even really mean when we're talking about scaling down?

- [Nandan] If you think about how AI has evolved so far, it has been very data and storage driven, right? So you actually gather more and more data, you build patterns out of them, right? And based off of that, you come up with inferences with intelligence to respond correctly. And as a result, there's more and more things you can find.

And once you start honing in the kinds of things you want to do, then you start building models that'll do those well and train them and then you use them for perhaps matching patterns, responding to stimuli, predicting things. Because if you think about intelligence in general, right? Especially what the brain does.

The brain is the- if you think about it, the most efficient learning and predictive inference engine, right? So it can take junk out of what you see. If you think about all pixels, that's what the eye sees, it's the brain that kind of then converts that into these are the physical objects, these that matter, these I should be worried about, these I shouldn't be worried about. And actually doing that processing at a very fast rate. If you think about it, that's a huge amount of data being processed very, very quickly. And so traditionally what AI has done is done that on the cloud because that's where you can handle that kind of data, that kind of performance.

But I think the ingenuity around is now, okay, now I can compress it. Now I can build models that are smaller and then I can start doing things closer to where the actual sensors are. So if you think about what edge is, it's basically saying actually they're moving intelligence closer to where it's needed, right?

And where you need to respond to a stimuli because the sensors are providing you stimulus, you are expected to respond to it, right? So that's the high level view of what edge means. We have been spoiled in some ways to say, okay, we are always gonna have this high bandwidth connection to cloud.

We have compute at our disposal at every point in time. So everything we need can go to cloud, get computed, come back immediately. But the amount of intelligence and the amount of compute now needed and the amount of bandwidth now needed is beginning to start, what should I say, limiting how you scale if everything goes to cloud, which is why you need more things moving closer to the edge.

- [Ryan] For somebody listening that out there who's thinking about moving things to the edge, or that computing power, that intelligence to the edge, what are the benefits that come with doing that as opposed to say having that done further away from the edge, like in the cloud, for instance.

- [Nandan] Yeah, that's a very fair question, right? It's all a question of economics, one, a question of security, two, and it's a question of timeliness, three, right? So if you look at- these are the three basic things that you care about. As you start thinking about sending things to cloud, you're talking about connectivity charges, you're talking about compute charges, you're talking about storage charges, and a much more generic heavyweight model does not understand the context of where that data is coming from often, right?

So the economics of it start getting pretty crazy. Second part is the timeliness of it. If you don't do it immediately close to where it is, you, the point- the response may be too late, right? So think about a car that is trying to process data, loses connectivity for a second, it's too late when it gets the data back.

And the third thing is privacy. There are things that often get sent that you don't want to be sent. If you're doing health data, yes, I'm gonna try to encrypt it, send it, which means more processing, one side or the- and the other, right? Privacy. All of those things start contributing to why you want things closer to the edge.

And finally, if you think about- put it all together, you need to understand that anything we do in terms of compute that has to scale adds a real tax in terms of the energy needed to compute it. All the costs that go into the overall side of it, right? Moving things closer to where they are, just like localization when you come and talk about logistics, just like- be local, by local. There's a reason for that. You can be more efficient closer to where things are rather than having everything go to a central location and come back out.

- [Nikolai] What would be the trade offs with edge AI? We just talked about the benefits, but what are you giving up compared to a bunch of GPUs?

- [Nandan] Edge AI, in general, I would say is going to be specific and special purpose for most part, rather than generic, right? GPUs are great at vector processing, a lot of parallel compute but hence also heavyweight, right? Power hungry, expensive, right? They will, by the way, edge is a pretty broad word, so I think we should qualify that.

Neil's smiling because he knows, right? You think about sensor edge, which is really, really close to the sensor itself, which is really small, very cost effective devices that may run on a CR2032 type cell, right? Then you talk about the embedded edge where you see a lot of microcontrollers today, right?

Doing, hey, I'm- I sense somebody coming in, then I wake up the bigger processor and then what we call network edge, which is sitting still on your side of the network, but it's a mini server at that point, right? And this is where you see NVIDIA's Jetson or a Qualcomm Snapdragon tech products playing.

So what you're giving up, at times, is the general purpose aspect potentially as you get closer, but that- you don't need that, right? If all you're doing is understanding a few phrases, you don't need a giant GPU telling you can do that, right? If you're just monitoring a health signal, let's say white will sign, you don't need that.

So it's a trade off between efficiency and general purpose. So if you get one device, you're not gonna be using it everywhere. Whereas let's say you took a what's today considered an edge server box or an edge AI box, you may be able to do different things on it, but again, you pay for it in terms of power, in terms of cost, et cetera.

- [Neil] You're bringing up an interesting point, Nandan, because 12 years ago, back when I was in the early days of Watson, we were focused on healthcare, and people were actually asking, we have a very specific use, I don't have to worry about cloud and all these other things because people thought we had built a machine, and we did, and we said we built a computer.

There's a difference. And- but they're basically saying why didn't build the chip, this kind of stuff. And we're trying to explain that the investment infrastructure hasn't really been made in 30, 40 years. It's really been a lot of software focus. So it was not original goal.

Things like neuromorphic chips and quantum computers were not there yet. Do you think given everything today and like some of the work you do with BrainChip, have we reached an inflection point where hardware is actually becoming the next frontier?

- [Nandan] That's a great question. So I'll step back and talk about eras of compute, if you will. So if you look at the 1960s, you were talking about vacuum tubes, and you were actually digitally pushing switches, right? Was that- when the punch cards came in, was that a big step? Yes. But were we there? No. When keyboards and screens came in, was that a big step? Huge. It changed the game quite a bit, right? Were we there? No. Then the touchscreens and gyros came into your smartphones, completely changed the game? Not quite. So I think what we're- my point being, I'm not trying to be evasive here, but we are taking a big step forward.

Is it completely there? No, but it's a big step forward as we get to neuromorphic on the edge. And finally, neuromorphic is just a mechanism for smart compute or intelligent compute. And the only good thing about it, and I'll be shocked for saying this, but the only good thing about it is it is actually much mimics the brain in how efficient it can do things, right?

But that is actually the calling card for any time when you wanna do very efficient compute.

- [Neil] That's why DARPA is, was it half a trillion dollars, half a billion dollars, I forget what it was, they've already invested in neuromorphic chip research?

- [Nandan] So it is, it is. I think and by the way, we have focused neuromorphic on the edge, right? But neuromorphic is equally useful at the network edge, in the cloud, everywhere. It's a basic way of thinking about it, no pun intended, right? How you solve the problem the way the brain actually processes it, right?

And we've done very well in compute, we are actually- we were mapping an intelligence problem on the compute we had today. This is rethinking it to say that's how it computes and we'll start thinking about how to solve different problems in a different way, and I think the last 20 plus years, there's been so much progress.

But there's a line of thought that says neuromorphic is still out there. So what BrainChip tried to do is to say take those principles, build out something that is very portable and scalable, and not have to wait for spiking neural nets, et cetera, to make it to the mainstream. We are actually mapping today's networks, CNNs, transformers, et cetera, onto hardware that understands how to compute like the brain, right? And that's a huge step in my view. We focused on the edge because we feel that's actually a great application area for scale, right? Because it's much closer to the sensor, and it helps us to start thinking in a distributed fashion. Neil, you seem like you've been through that type of process.

- [Neil] I've had a similar conversation many times. It's always the juggle with technology, right? It's always the hardware, software, and people are always just like why can't you guys do more with AI? We don't have the computing power.

- [Nandan] And this is naturally- one of the key challenges that most new technologies have is, hey, I've got a great technology. Use it. And like, how do I use it? Can I actually take it to market fast? Can I map it to what I have today? How do I transition? Those are really compelling problems to solve.

If you do that, you succeed. And what we've taken an approach to is to say, okay, you don't have to rethink how you write your CNN. You can take this as it runs. We'll provide you the tools to map it into our hardware, which is much more efficient. Now, next time around, if you think a little bit differently, you can make it that much more, but the first step itself gives you an order of two of magnitude.

If you start thinking spiking, then you're talking a couple more. So it's future proofed in that regard. We'll see how it works, but I think that's really the small set of dominoes that you've set in motion to start seeing the broader market. Take that on as well.

- [Ryan] Can you just for our audience's benefit here just explain in layman terms what it means for neuromorphic computing, how it's inspired by the brain, what this actually is. I know there's some of our audience that's probably fascinated by what we're talking about, but may not exactly understand what it is and what's so special about it.

- [Nandan] Today what we see is AI is a lot of parallel compute. And the data that goes- the inputs that go can be sparse, as they say, right? Not everything has to compute, and you can use that as intelligently as you can to minimize the amount of compute that needs to be done or the performance that needs to be generated for it.

Neuromorphic thinks about it slightly differently. It says, again, let me think how a neuron is built and how it connects and how it actually competes beyond that. So in general, neuromorphic computing is looking at only driven by what are called spikes, right? That's how in the brain things move around.

You have a spike that triggers neurons that are connected to it. Some of them trigger, some of them don't, and you actually only progress down the path that you need to progress. Spiking behavior is the same way, or what we call it event based, right? So only when an event happens do you compute.

And once you compute, you know whether the next neuron, if you will, has to trigger or not, and it triggers the next. So what that does is it by orders of magnitude reduces the amount of computation required in terms of wasted computation is thrown away, right? Effectively, it mimics the brain and how it works because it's extremely efficient.

People- if you measure it, the brain takes about 20 watts of equivalent power to compute substantially more than most supercomputers do. So we're trying to use that model.

- [Ryan] So does- do you see the neuromorphic chips democratizing AI hardware and computing. Like how does it compare in terms of costs and what are the trade offs and things like that?

- [Nandan] The other approach that BrainChip's taken and which is not surprising is that we believe that AI is a compute of- a component of every compute problem. And as you've probably seen, the transition from traditional, hey, there's a compute block versus a system on a chip, as it's called. You saw that thing come in with embedded controllers like microcontrollers. Then you started seeing that go through a system on a chip for application processors for phones. Majority of the compute today that's out in the world, 90 plus percent is system on a chip or embedded type compute.

And so what we do is we integrate, license the IP to integrate into somebody's system on a chip. So what that does is people can tune it to the exact amount of compute they need, right? And then rather than having a general one size fits all type solution that maybe often overpowered and over costly for those applications.

So what does that enable? So for example, you could think of a like an embedable or wearable health monitor that you could create that could run on a CR2032 type thing forever, as in, okay, I overstate it, for months. Or it could be potentially even smaller and could be harvest- energy harvesting type solutions.

Those are possible. You could see a lot of these portable devices now going into the field that have maybe battery powered larger batteries, of course, but can do much more intelligent work on the field whether it is object detection and understanding, they can actually understand, let's say, hey, I took an ECG remotely before I send it to the cloud or to the clinic.

Did I take the right readings? So that's an example you could see. Obviously when you talk about security surveillance, it's very clear that you can do that. You're talking about devices that don't need heavy duty packaging or fans. The economics of silicon works a certain way, right?

It's not just the size of the silicon that's a cost. Actually, the packaging carries a lot of that cost. So the hotter the silicon gets the more you need to pay to package it. And so that comes into the economics of what kinds of devices can you go through? Can this go into all types of field devices?

And is it cost effective for that? Can I get a $2 chip instead of a hundred dollar chip? Those are the kinds of things that we enable with IP that is configurable, portable, and integrated into SOC solutions.

- [Neil] What do you see as the long-term benefits for businesses then?

- [Nandan] There's clearly lots of benefits. There's initial investment, naturally. Everybody needs to get savvy with what they can utilize the AI for. And so from a business benefit standpoint, there's scale, right? So today I think you're starting to bottleneck on costs at cloud, storage costs at cloud, and connectivity costs at cloud.

If you start using devices that are connected to cloud, but not constantly reliant on it for all the compute that is needed and large amounts of data going back and forth, suddenly those devices become much more cost effective, right? And once those devices become cost effective, the services become cost effective, and then you can actually scale to more clients, more types of applications and go from there.

So I'll give you an example, right? So there was a lawnmower with weed killing capability which had a giant, I'm not gonna say who's but multi hundred dollar system that helped identify weeds and shoot it, right? So you can shoot the right weed collected- that may work for your premium lawnmower. It's not gonna scale down to the more consumer friendly price points, right? Now if you did instead used something like a BrainChip along with a smaller AP and an SOC and now instead of a general purpose solution, you have a special purpose solution, right?

It suddenly changes the economics of how far it can go, right? So you can scale your offering to a broader clientele. And for a lot of these today, the economy is moving from a device economy to a service economy. So everybody's not just selling the device, but there's a service that goes with it. The more people that are on your service, the better your chance of success.

- [Neil] Volume game, right? You wanna generate the transaction.

- [Nandan] Exactly.

- [Nikolai] I have I have a question about the implications of edge AI. You can imagine- I guess this is kind of existential, but you can imagine the AIs of the future are going to have a hand in everything. In your lawnmower, in your smartwatch, in everything because of edge AI. So, it's like AI is getting like real-time data on a global scale at a very granular level. So I'm wondering what are the benefits and also the risks of that. Should AI- should all that data be centralized somewhere? Or should AI start acting more autonomously based on all that data?

- [Nandan] Yeah. Yeah. This is getting into the philosophical side, which I think it's interesting. The benefits of AI moving, actually decentralized, are clear, right? So one is that you get better chances to control critical data or sensitive data, right? You don't want data that's close to your- let's say your health records, et cetera, to be blasted by some service that you didn't know of and they get tapped, right? Your personal information. Certainly privacy, security are important. There are clear benefits on, hey, if you look at fitness, and I have to see the statistic to see if it really helped, but Fitbits and fitness wearables started helping people understand how they're behaving, how they're sleeping. In fact when the first Apple Watch came out with the heart monitoring capability of- I actually know of a colleague of mine who found out an arrhythmia type. It didn't tell him that, but he saw something enough to go to the doctor saying, is this right? That led to a preventative health event, right?

And I think a lot of the benefits are going to be in preventative, whether it is manufacturing, whether it's healthcare, whether it is automotive directions. I'm giving you kind of simple answer, but there is going to be a lot of benefit for learning, and as we get more ensconsed into our social network and our phones, et cetera, funnily enough, we need the devices to start keeping us honest and keeping us aware.

So this is actually a part of it. The negatives, I think can be, there are many, right? So like any great technology, there is going to be pluses and minuses. Can it actually reduce the amount of jobs that people need to do that- which is the constant discussion there. I think there will be more occupations and capabilities that humans will find to do that make them economically capable.

So I'm less worried about that. Can it be misused? Sure. Like anything, anybody that has access to lots of data, especially personal data, it can be misused, which is another reason why distributed approaches, I think, help with minimizing that problem. Decentralization, in my view, helps minimize that problem.

- [Ryan] I wanted to shift our conversation just a bit to a couple topics I think would be interesting to get your perspective on. A little higher level topics. But from your perspective, where we are right now as an industry, given all the different technologies available to us, where- what do you see from your all's side as the top applications of like true AI technology?

What are- we obviously know there are limitations out there with AI technology at times, but what are the top applications you're seeing AI provide to the world?

- [Nandan] Obviously security, surveillance, all those kinds of things have been used already. You're beginning to see this rush towards autonomous driving and autonomy so to speak becoming- I mean there's a lot of value there, right? Interestingly enough, autonomy could help reduce the amount of road fatalities substantially.

Interestingly enough, freeing up more time for you to think about other things and more capabilities. But I thought there was one thing that I would say was coming together much more interestingly. If you look at the costs of education and healthcare, they've been the ones that have been spiraling quite out of control in a lot of ways. And one staggering number that we talked about when we launched our second generation was the CDC statement that the cost of lost productivity for people showing up due to preventative chronic health conditions was $1.1 trillion dollars. That's the lost productivity, meaning not showing up to work and hence lost productivity.

It's not the cost of fixing those problems. So if you can start making people's lives healthier earlier, preventing all those things that would lead to a condition that then becomes dependent on constant need for service- health services, right? That would be a huge benefit that can be elicited from it.

How do we actually administer healthcare in remote areas becomes much easier through things like AI. So if I were to take a positive view on what AI can help us with is our own health.

- [Ryan] That's interesting. We actually just had a discussion with another guest earlier today who's focused on AI and healthcare. Being able to better understand people's body types and identify things that are right, wrong, and mixing that in with all different types of data. So super fascinating to hear you back that up as well.

- [Nandan] In fact, if you notice, most people over the last five to 10 years, leave TikTok aside, have been focusing on apps that help them meditate and reduce their sugar intakes or get the right exercises, et cetera, right? So people do care about how they feel, and that they're healthy.

AI helps them do that better.

- [Neil] You think that's because of the assistant model that there's a nonjudgmental AI out there giving you advice or seeing that you're doing that exercise not quite right, move that elbow up a little bit more. It gives that little nudge.

- [Nandan] I got a degree in computer science, not psychology, so I can't speak to that, but jokes apart, no I do think as from a personal level, taking courses at your pace or doing things at your pace is- there is value to that. And more importantly, I have a traveling kind of job, right? So I don't- I can't guarantee that I'll have my instructor with me at any point in time.

These things do help. I think a lot of that is unlocking the ability to be independent and instant gratification, if you will, on the tools that you need.

- [Neil] I mean, you're basically, I think, alluding to what the holy grail in AI is is the personal concierge that knows you as well as you know yourself. I know I reference this a lot, but the White Christmas episode of Black Mirror is a good example of that where they take your brain in grams, create the AI assistant, so it can actually anticipate what you need.

They- I feel like that's what people want, but you know that cloud and storage isn't enough for that. We need to have some of that logic and processing power built into the chip itself.

- [Nandan] Indeed and that's the personalization aspect, the customization aspect that you need. You think about AI as a movement if you will, right? Has been about making a- you being a part of a bigger intelligence and that bigger intelligence making you better, right? That's the philosophical goal of this.

And I'm hopeful that we can get closer towards that.

- [Ryan] Let me ask you, one thing I was thinking about is as we're talking about these applications, not obviously healthcare we're leading with, but just as the technology continues to evolve, where do you think, and obviously we- and we're tied to the IoT side, not for this podcast, but like our experience in the past on my side is we talk about how IoT, enterprise IoT is a really big driver for enterprise AI because of the ability for it to collect data, right?

And that's where the AI models can get the data from. But as we look into the future of AI data when more businesses are adopting AI applications and solutions, where will that data come from? I know it will come from a variety of different places for sure, but where- what do you see changing when it comes to the acquisition of data for these models as technologies and enterprises start to adopt more of these solutions and tools?

- [Nandan] At the simplest point, data comes from sensors. So if you think about it, the- what is- what are you sensing, whether it's chemicals, whether it's action, motion, vision. They all come from sensors, right? So if you think about where the big enterprise IoT solutions for AI come in, there's a clear value in, let's say, how do I understand vibration?

How do I understand maybe chemical concentration in my manufacturing plant. How can I ensure that it- I can avoid a negative condition? How can I stop it from being a production stop? So clearly that is using all the various sensory data to predict and prevent negative conditions.

You could see the same thing start coming in from- we talked about the five human senses, right? So we showed how BrainChip technology can be sensor agnostic by showing how it can actually use sensors to tell you what type of beer you're drinking, right? I don't need help with that.

But still you can do things like that. You can do olfactory, you can do vision, you can do touch sense, et cetera. So I think data will start coming from all types of sensory sources. Is the dryness of the leaf on my farm indicative of anything that I should prevent or be worried about?

And if so, what should I be doing? So there- I think there'll be sensory data coming from all types of things that kind of bring into it. Obviously things like financial data, spending data, where I spend my time, those things are already built into your devices. I'm looking at other things that will crop into it.

- [Nikolai] This is just a random comment, but if you were to channel all this data into a central AI, it would essentially be like an all seeing AI, that was like a global brain.

- [Nandan] Fascinating. Yes.

- [Neil] Just so our listeners don't freak out, that means that everybody's chips, all different makers, all different big tech companies would have to consolidate all their stuff together. That's never gonna happen.

- [Nandan] And I do think that's- the movement always goes back and forth, right? So you look at every industry it goes to centralization, decentralization curves. I think it's a natural thing now, both from an economic perspective, security, privacy perspective that we're moving away from the centralization model.

So I'd worry a little less about that. Hopefully. And we go towards a faster, quicker response time, secure AI.

- [Ryan] One of the last things I wanna ask you before we wrap up here is just for audience out there listening thinking about bringing AI into their business or the business of their customers, what are those opportunities that are going to continue to exist and probably come from as this continues to grow to businesses that are adopting AI?

What do those future opportunities look like in your mind?

- [Nandan] Let me throw a couple of examples your way, right? And then there's many more obviously that go through that. So one is that every time you make a transaction or an interaction easier, it's good for your business, right? And so the more human the interaction, and by that I don't mean, oh, I need a- any kind of- if I can talk to something, and it understands it and helps me move along with it, it helps. Rather than if I have to go through 15 different menus to choose from things, right? So there is the human interaction aspect that certainly starts helping businesses, which is why you started seeing lots of bots come about, right? Every website now has a bot that will interact with you.

Not all of it is AI, but it can be, right? The second piece around that is understanding and developing whatever makes your business better. What is the feedback from the customer? Can I understand it? How can I help my product be better?

- [Ryan] That's great. This is a fantastic conversation. Neil, Nikolai, anything- any last comments or questions before we wrap up?

- [Neil] I think it's been a fascinating insight into the hardware side and what's going on with AI chips and neuromorphics.

- [Nandan] Well, thank you. I think this is a great forum. I really appreciated the discussion and hope to see more of this type of discussion ongoing where we are actually discussing specifics of what's happening at the edge on IoT because if you go 15 years ago, we were just talking about what is IoT, right?

And now we're talking about AIoT. It's a great step in a decade.

- [Ryan] For sure. So can you tell our audience where if, as they're listening to this, if they wanna follow up with any questions, learn more about BrainChip and what you have going on, what's the best way to do that?

- [Nandan] So BrainChip, easy. The name's quite simple. brainchip.com, brainchip.ai. Get to our website. Try it out. We have YouTube channels which I can provide you with to connect to. If you search for BrainChip on YouTube, you'll find it. And then we are actually available. We have a podcast network.

Also we talk to a leading- just like you've set up here, to go through what are the trends, what's capabilities. We have all the resources to help you understand what we're doing, what the industry's doing, and how to interact with us. So again, brainchip.com

- [Ryan] Thank you so much for your time. Really appreciate it and excited to get this out to our audience and hopefully speak in the future.

- [Nandan] Thanks again, Ryan. Thanks Nikolai. Thanks Neil. This has been great. Thanks for the opportunity.