BrainChip Holdings Ltd (BRN) Earnings Call Transcript & Summary

This is the BrainChip podcast, hear from our thought leaders about neuromorphic computing, beneficial AI and how BrainChip's Akida is pushing AI to the edge. This podcast is a place for investors, practitioners and anyone interested in the future of AI.

Hi, all. I'm Rob Telson, Vice President of ecosystem and partnerships at BrainChip. Welcome, and thank you for joining our latest episode of our BrainChip podcast series. These events are structured to provide current and future investors and those interested in AI. And the brain chip technology, a path to better understand who we are, what we are doing and where we are going. If you've not listened to any of our podcast, please go to our website at www.brainchip.com, go to our media tab, where you will find our podcast. You can also listen to any of these podcasts on your favorite podcast platform or please go to our YouTube channel at BrainChip Inc. and find all of our podcasts and additional media. It's been a while since we've taken this approach. It's what I call an inside looking out approach. Traditionally, we do outside looking in. But on our podcast series, so today, we have the pleasure of having BrainChip's own Peter Van der Made, Founder and Chief Technology Officer; and Nandan Nayampally, Chief Marketing Officer on our podcast. Peter Nandan, welcome to the BrainChip podcast.

Happy to be here. Thank you.

Yes. Thanks for having us on, Rob.

Yes, I got to tell you, I'm very excited about today's podcast. The intention today is that this is going to be like a fireside chat, more of a roundtable, but we're really going to be able to get into the thought leadership. So being here with both Peter and Nandan means a lot of valuable insight into BrainChip, what we have accomplished, where we are going and the future from their perspective on the world of AI. So I truly admire the work that Peter has put into driving BrainChip's architecture. His efforts and vision have been the cornerstone of BrainChip and personally thank you, Peter. And Nandan brings very broad industry experience with a deep product marketing background, decades of IP experience at ARM and other companies and most importantly, a very strong AI background as well. So for our listeners, I've had the opportunity to work with both of these gentlemen and can truly attest we are in good hands. And these are the things that get me excited.

So let's get started. Let's start with -- I'm going to try to do this kind of as the host of this discussion here. And I'll start with Nandan. Let's focus on edge-based AI and what are the key drivers and some of the problems that you feel currently need to be addressed in today's AI market?

Yes. Thanks, Rob. So let me start with a common view, right? As cloud has grown and things like Internet of Things have grown along with it, the assumption always was that there is a dumb device at the corner, most of the intelligence and the computations done at the cloud. It creates lots of challenges. And what you've seen over the last decade or so is that intelligent compute, power-efficient compute is moving closer to the end device to the end user and only the right amount of actual computation is getting done on the cloud. Now AI follows a very similar process. The overall intelligence grows as all these local intelligences contribute to it. But unless you have local intelligence, you have a bit more stunted global intelligence right? And let's talk about the commercial aspects of it. The cost of compute at the cloud is growing rapidly. The more the data gets thrown in and more complex things that need to be done or applications that need to be done, the bigger the bill at the back end. And especially things like storage, nearly 2/3 of the cost of cloud compute is in the storage aspect. So a natural side effect as a part of it is the network congestion. But more importantly, what you need to do at the Edge in terms of how responsive is it? Is it fast enough to give you the experience. If I was in a zone that could not connect to cloud, would I be affected badly if I was driving and I needed immediate reaction, and it had to be done in the cloud, you couldn't do it. You see similar things in industrial, you similar things on your own personal devices. Right? So effectively, that is what is driving the need for Edge AI. Now you add to it the fact that a lot of this data is now sensitive. You want more intelligence at our device, both to protect sensitive data, private data, but also reduce the amount of data that goes back and forth. So these are some of the key challenges that are driving the need for Edge AI.

Thanks, Nandan, appreciate it. Peter, what do you see as -- from your perspective, some of the key challenges as well that currently need to be addressed?

Yes. As Nandan already mentioned, the traditional way of doing things is to go up to the cloud and build computing in the cloud, that presents certain problems now. One of the problems that are not obvious is the enormous power consumption of the data centers around the world, where the total power consumption of those data centers is larger than the entire United Kingdom is using over a year. So we have a solution to this problem in Akida in that we have a very low power consumption device with a terminal footprint that is extremely small -- moving devices to the edge of the -- close to the application. You can say -- we have calculated this something like 97% of their power. So Akida we run of 3% of the power that currently is being consumed if we were to move all the applications to the edge. So that's the very strong driver, especially in the current climate where we see climate change and all the problems with more environment well that we need to need to address. So moving devices like Akida on to the cloud -- sorry, until the Edge from moving things away from the cloud, it means that we well impacted that the problem.

Yes. just building off of what you said, Peter, for our listeners and just let's take that and digest that. We've talked about before that we're at the forefront of this revolution. And what you've just heard Peter say is basically the means to an end is that eventually, the amount of compute that's going on in the world today, we can't sustain it. And it's going to take technologies and advanced architecture like what Peter and his team have started and done with Akida and what we're building on to support the compute as we continue to evolve -- and so this is -- what you've just heard very clear is that, hey, this is a big problem. It needs to be addressed. And it gets addressed by companies like brain chip that develop products and then introduce new products and build upon the technology as we evolve. Before I move on, is there anything else that you guys want to add before I go to the next question?

Yes. One of the important factor is that the Akida offers high performance at very low power consumption. And that's key to people want high performance at EH. That's a very important part of our value proposition is that we can offer the performance and yet a very low-power footprint, low terminal footprint.

And yes, when we're thinking about thermal footprint and power, I think what you get there is that now you combine that, you combine the performance factor with the ability to really manage that power and where we can take the technology expands dramatically. And what you'll see -- I'm going to move on to the next question because it really ties into what we just talked about guys. And basically, let's look at what we have today. And what we have today is we've developed the IP that you can design into chips that enable you to build the AI engine into your device and support all of this compute moving forward. And we've done that based off of a product that we call the Akida 1000.-And so we offer that product as silicon as development systems and also in boards to enable companies to start building prototypes and move forward at a very fast pace as they continue to design out their AI strategy. But when we think about it, we've had shareholders and we've had people very interested in BrainChip asked the simple questions of, okay, well, if I was going to take a keto today, what are the perfect markets, what type of products would it go into that it fits in very well right now. So Peter, just building off of what we've talked about, if you are a customer, a BrainChip, what would be that -- to you the perfect customer that will be taking our technology today.

There's a number of perfect customers. It's not just a perfect customer you see the applications in consumer markets where you have products that you can speak to and [indiscernible] language, the professional recognition in doorbells, in access control. In commercial applications, in robotics, there are so many applications for Akida that it's difficult to pick any specific applications that's the ideal application.

Yes. And what's really unique is that we have a very dynamic sales force that we're continuing to build out and scale. It's grown dramatically over the last few months by the leadership of Chris Stevens. And what I want to emphasize beyond that is that we've also started to build out on the ecosystem and partnership side. But all of that is enabling us, as Peter has kind of highlighted is that we have a broad spectrum of where the technology is going to fit very well in a very short amount of time. But Nandan, do you want to build on that? What's your mindset?

No, I think that's a really good starting point, right? So Akida 1000 was designed to enable partners to prototype. And actually, BrainChip has morphed our business into really recognizing the ability to build our engine into different system-on-chip solutions. Right? And the key benefit of this is this gets combined with other compute you may have other intelligence you may have into solutions that fit particular markets. So the Akida technology is much more broad-based than just the AKD1000. Having said that, it has actually helped us blaze the trail in a lot of ways. You can see automotive, in fact, some of the big announcements that we have seen are in how does a very compact AKD1000-based solution help with in-cabin experience, more intuitive human interaction with the machine and especially in the automotive world, where we are moving closer and closer to partially electric or fully electric vehicles. Any energy drain is a challenge -- any reduction on data across the system, communication across the system that you can manage is a big benefit. You see similar situations in industrial. What we find AKD1000 setting us up to do really well is to get closer to the sensor, bring more intelligence to the very edge, reduce the amount of computation that gets further up the network, reduce the amount of data that needs to go further up the network, protect this data going further up the network. So you end up seeing interesting solutions for any kind of human machine interaction that is more intuitive, whether it be in automotive, whether it be industrial, whether it be in consumer, whether it be in health care. In fact, -- some of this technology will lend itself really well in future devices that can be embedded and run on batteries for years or energy harvesting. And AKD 1000 helps us put that platform in place for people to develop those prototypes, develop those innovations, develop those models that can actually be taken to market.

Yes. That's great. And Nandan, thank you for explaining it that way. That you were spot on. One of the things we just announced, very exciting, a lot of people were interested in what's next from branch, where are we going, how are we getting there? What are we doing? And so we announced AKD1500, which is basically an 8 node piece of IP, and we're validating that using our partner, GLOBALFOUNDRIES. And we've done that because what IP is like, it's like a cookie recipe. And the important thing about that cookie recipe is you want to make sure it works the same. It tastes the same in any environment that you bake it in and you build it in. And so we've partnered with GLOBALFOUNDRIES in a 22-nanometer process. So we're making it smaller, much more efficient from a power standpoint, much more effective, but much more powerful as well. And so when you look at AKD1500, again, we're building off the same theme of there's a lot of areas in which we can take this technology as we continue to evolve. But in specific, Peter, are there any other -- any specific areas where you think AKD1500 is going to thrive?

Yes. AKD 1500 is a much smaller footprint than Akida 1000. It doesn't have the CPU on board and other bits that a customer may not need. It's very suited for consumer applications and sensor applications, industrial sensor applications. But these are the main markets that I'll see application or on Akida 1500. Yes.

And just building on what Peter said talking about consumer-based applications and industrial-based applications, for our listeners, we expect those 2 areas from a market standpoint to be very high volume. And to us, volume is very important. Nandan, do you want to add anything?

Yes. I think let me start back with the philosophy that you started with, right? We have proven it out in 28-nanometer, which is where the AKD 1000 was. Now moving to 202-nanometer FD-SOI silicon on insulator process, it is going towards much more of these low leakage low-power applications, as you say. But it highlights a couple of other things. We talk about scalability. And one word we haven't yet mentioned about Akida. It's based on neuromorphic principles. It's an event-based architecture that draws from how the brain works and the assumption in the industry is that neuromorphic means analog because that's how our brain works. What's cool about Akida is it's designed as a digital, a fully digital architecture and hence, very portable across different foundries. And what -- our AKD 1500 does show how portable it is across different process types, different process technologies. Now the second thing AKD1500 does is focus on what you talked about as volume markets, but really truly making it an accelerator, right? It's designed to be a processor or an accelerator. It does not need its own CPU. It can be created into a module with an MCU or added as an accelerator for partners that want to build cards out of it. It's a great design to prove that out. But the second piece is one of our key strengths is for at sensor intelligence, especially for feed forward networks that fit. We don't even need a CPU. So we can actually reduce the overall [ bomb ] cost for intelligent sensors by just having this accelerator along close to a sensor, it manages its own access to models to wait, does its own computation. So AK-1500 is a great vehicle to demonstrate what is achievable, especially at the very edge. Now the other thing you mentioned is that it's an 8 node smaller footprint platform versus an AKD1000, which is a 20 node platform. One of the key benefits of the Akida architecture is the ability to run what we call multi-pass operation. And what that means is if you're constrained in footprint and you still want to learn or run larger models, we have the ability to do that on its own, completely transparent to the user and still achieve the kinds of metrics that we need. And this is another way to demonstrate that. And actually, a great way to showcase how we can start with a particular configuration that you have and future proof for heavier applications coming down the road.

Thanks, Nandan. And one of the things I really appreciate about the way that Nandan paints the picture and communicates. He likes to leverage the word intelligence -- and sometimes I say compute, but he's really right. This is about driving the intelligence and making devices, whether they be on the shop floor, in our home, in a vehicle or a lot of medical devices, making them more intelligent. And that's what we're driving forward with our technology. Let's have some fun. Let's start to look at the future a little bit. And Peter, this is where we get excited because this is the stuff that you focus on, on a daily basis. But let's look at in the short term, we'll call that the next 3 years. So when you look at it in the next 3 years, tell us what -- from your crystal ball, what do you see?

Well, yes, we are actively researching new network architectures for the future and here in Perth. The entire team here is dedicated to finding better ways of doing things. The -- what we see for the future is that we will be expanding our market beyond what is traditionally the edge. So we will take a larger part of what's currently being done in the cloud. And I'm particularly thinking about like self-driving cars, robotics and also the home market -- home automation market. For that reason, we are studying the neurology of the upper layers of the human brain, the cortex is where all our conscious spots are taking place and high-level processing happens. And we're implementing those architectures in -- within the neuromorphic core. I just want to explain a bit more about the neuromorphic core because purely neuromorphic has a bit of an -- and name to it that people think that they think analog, as Nandan mentioned, that's not where we are. We are doing everything digital. We are also ramping that neuromorphic core with convolutional functions so that convolutional networks can run very efficiently on the neuromorphic core. The neuromorphic core is extremely low power. The overlaying convolutional layer, convolution shell that sits on top of the neuromorphic core enables us to interface to existing applications trending, but also we have built of continuous learning so that we can use the features that are in the convolutional layers and then use the neuromorphic core to learn new objects. So that's -- all of those things are very exciting things for the future. And we're expanding on that in that we're using the same type of learning within the cortical architecture, which means we make Akida smarter and smarter.

This is where it gets fun. Whenever we're in planning sessions and I get to hear a little bit more about what Peter and the team were doing in Perth, that's again another reason why I just shake my head and say, yes, we are on to something very special here. Nandan, do you want to add anything when you look 3 years out from now?

I was going to poke Peter a little bit and say, "Hey, we do learning today, right? So AKD1000, AKD1500 already have the ability to do on chip learning, one shop few shot. And in fact, some of the videos that illustrate how actually capable that learning algorithm is or learning capability is on today's platforms should be quite kind of an eye opener to what is possible when Peter gets to delivering the things he's talking about right? So I think let me kind of step back one second and say, right, we're talking about extremely efficient devices at the edge that are much more intelligent, right? Now this is a spectrum. This continues. There may be step functions in the way we improve our architecture, and you'll see lots coming out as we go along. It's not just a 3-year step function process. We're incrementally improving it. You can think of, as you say, more and more complex workloads being done on the device. So can you go to a trail cam and not be connected to the cloud and fully recognize your terrain, -- can you recognize the win life? Can you recognize the fan, and the fauna Flora, -- those are kinds of things that you may be able to see with specialized applications using intelligent low-power implementations of engines like Akida. Could you actually proactively monitor health warning doctors ahead of a capability or an issue that a person has. Now again, remember, when you're talking about health care, most of the signs are guidelines, right? Each person is very different. So if you have devices on you that are very generic, -- you're not getting the full benefit that you need for a device that learns more about you and customizes itself. And once, let's say, an embedded device is installed in you, you can't do the operation every time to take it out. So the ability for that to future proof to improvements in algorithms, but more importantly, customization that you need is a very, very interesting scenario. We talked about the AI labs announcement we did earlier this week. Now according to various studies, unplanned shutdowns of production lines are $50 billion costing $50 billion to manufacturers yearly. Can you envision the amount of productivity gain if you were solving those problems ahead of time?

Absolutely.

What innovation could be had when you keep that all possible. So I think the interesting thing you will see going forward is the ability to predict problems based on all your local data and sold them ahead of time so you can actually get the benefit of this intelligence.

Yes. Thanks, Nandan, and that's great. Okay. We're going to look further out. I mean it just so happens it 7 years from now, we'll be at 2030. Wow. Let's talk about 2030. So Peter, when you think of 2030, you think of BrainChip and you think of brain chips Akida and the evolution of what we're doing for our listeners in a very short amount of time, what would you want to convey to them?

Yes. 7 years from now would be our target for what we call Akida at 10, which is a device that doesn't just learn to recognize new objects or other new phases or whatever. But it learns to interact with the world the same way as human being interact. Learning is a very interesting part of research in that it is the gateway to intelligence. Children are not born with full knowledge. They need a long time to learn. We are thinking of future artificial intelligence in the same way, and it's born with a sale standards of knowledge, but it expands on that mass have learned so self-driving cars learns to drive better over time machines in this industry learn the task better over time. This is a holy grail of artificial intelligence is what we're working towards is artificial general intelligence.

I'm going to ask Peter, are you predicting AGI by 2030, Peter? Do we have the singularity by then?

We will -- I don't like the word in the library because humans are quite unique. We are very creative. We are dream. I don't think we've built machines that have the same capability by 2030. I'm not sure if you want to build machines that can envision different futures for themselves. What we want is machines that are intelligent enough to learn how to drive a car or learn how to operate the machine, learn how to make pancakes for breakfast. This is the sort of limitations that I would like to put on intelligent machines that we don't get into like a Hollywood scenario of robots taking over the world. I don't -- that's the reason why I don't like the idea of a singularity and people who look beyond the single art artificial super intelligence, humans are quite unique. We have EUR 86 billion neurons in our brain, we can combine things that seem totally unrelated to come up with new solutions, we don't expect to be there by 2030. And what we will develop is safe and beneficial AI.

Yes. Thank you. And we're going to wrap this up, but I think that the theme that we at BrainChip want to build on and what Peter just highlighted there is learning is the gateway to intelligence. And one of the things that we do very well is we learn. And we've just started down this path. So all right. nuts have some fun. So this is unique. Again, we're doing the fireside chat here. But for our listeners, they're used to me asking this question at the end. And Nandan, I'm going to start with you. It's a mainstay on our podcast. And we closed with our get to no question, which is if you could be one superhero, who would it be and why?

So with due apologies to Marvel comics, My Super Hero actually is my dad especially because apart from being able to do anything, we're able to handle anything the patience to deal with me growing up. But especially for the last 4 years, he's been dealing with his cancer very mutilating and debilitating, but has been the epitome of positivity. So like everything, right, you want to emulate a superhero, that's who I would emulate.

Yes, that's special. Peter, who would be your superhero and why?

Well, I've had a super here for some time. I think my super hero is not Marvel or anyone is a real person. It's Albert Einstein for the simple reason of how we managed to think outside of the box and come up with solutions that are -- that nobody realized we're there.

That's a good one. That's a real good one. So guys, I really appreciate you taking the time, and thank you. So Peter and Nandan, I want to thank you for spending time with us and providing your insight and feedback today. It's truly appreciated. For our listeners on behalf of the BrainChip team, we want to thank all of you as well as our customers and our future customers, our employees, our investors, our analysts and everyone interested in learning more about AI and most importantly, about BrainChip, we truly appreciate all of your passion and support. I promise you this, our podcast series will continue. And in 2023, we'll be doing some unique things very similar to this podcast and trying new things out. And until our next podcast, we wish everyone to stay healthy, stay happy and most importantly, stay out of trouble.

Thanks for listening to the brain chip podcast, please remember to rate and review on your favorite podcast platform.