Catching up with Brainchip CEO, Sean Hehir

 

BrainChip (ASX:BRN) interview

We caught up with Sean Hehir, CEO of BrainChip (ASX:BRN). Sean brought us up to speed on the 2nd generation Akida platform and the traction that has been getting recently. He also spoke about BRN’s capital structure and upcoming milestones that investors can look forward to.

Check out our recent investor webinar on BrainChip here!

 

Disclosure: Stocks Down Under/Pitt Street Research directors own BRN shares.

 

What are the Best ASX Stocks to invest in right now?

Check our ASX buy/sell tips

 

TRANSCRIPTION

 

Marc: Hello. And welcome to Stocks Down Under. My name is Marc Kennis, I’m one of the co-founders. And today we’re joined by the CEO of BrainChip, Sean Hehir. Morning Sean.

Sean: Morning, Marc.

Marc: We’ve spoken before about 18 months ago I think, about BrainChip obviously, but for people that don’t know BrainChip just talk us at a high level through what the company does.

Sean: Sure. What we do at BrainChip is we create intellectual property for people that are going to create custom chips that can do AI inference on the edge. When we say the edge that means outside the data center. And there’s a big movement in the industry to go away from everything in the data center to inference outside the data center.

So, things whether that’s automobiles, medical devices, hearing aids, things like that, industrial applications. So, we create intellectual property for people that want to build custom chips. Specifically, what we do, we have a way of doing this which is called neuromorphic. We’re the first company in the market to commercialize a neuromorphic approach.

Marc: And in terms of commercializing the technology, can you talk us through the model there? How does that work?

Sean: Sure. Our model is very simple like all licensing and IP companies. We sell a license originally, we charge some support in engineering for people who need help building their chips. And when the chips come off and are created we collect royalties. It’s a business model that over time creates a very attractive P&L.

Marc: And recently you announced the second generation of Akida, which is the actual product. Tell us how that’s different from the first generation and what it allows customers and prospects to do that they couldn’t do with the first generation.

Sean: Sure. I mean, it’s a wonderful question, something I could talk about all day. If you look at my comments I made at this last year’s AGM, Generation 1 as we call it the first release of IP is a world-class product. However, it’s very narrow in its functionality or very narrow in its use case.

So, we have some customers and prospects there. But when I talk about the pipeline later on the majority of [inaudible 00:02:04] interested in Generation 2 and why. Some of the things we did in Generation 2 were at the customer request meaning, “We want certain things that you don’t do in there.” Specifically, a lot of the industry is on 8-bit support and we were on 4-bit support. So, now we do 8, 4 and 2. So, that’s a big thing.

Some customers just say, “Hey, my models are on 8-bit. I wanna keep them in 8-bit.” So we did that. We included long-range skip connections, which is something that the industry wants.

Marc: What’s a long-range skip connection?

Sean: What is a long-range skip connection?

Marc: Mm-hmm.

Sean: It’s a term when you talk about the ability to do certain things with models without getting overly technical. But it’s a very common way to do things and we needed to include that. We support PyTorch, ONNX, PyTorch. Most of the models in today’s world are created in two major frameworks, TensorFlow or PyTorch, and now we support ONNX, PyTorch, which is critical. So, if anybody has a model it’s easy to put it onto Akida.

So, those are the things that kind of we’ve said, “Okay, the market spoke to us, customers spoke to us, said, ‘Please include these functionalities.'” So we’ve done that. Secondly, then, we’ve included some breakthrough kind of algorithm called TENNS, which is the ability to do things with the element of time called spatio-temporal neural networks.

When you read some of the materials, the breakthroughs are incredible, the ability to do much more with less, lower power consumptions, less parameters, and outstanding performance. So, we’re seeing a lot of interest in the pipeline for TENNS. And we’ve also included something called an encoder for vision transformers. So, that’s a lot of things that people can do that they couldn’t do before.

Marc: And so, if we bring that down to what we can see in everyday products, for instance, everyday functionality, or bring it really down to, say, products, where will we see Akida in a couple of years and what type of products? And looking at those products what does Akida allow that particular product to do that wasn’t available before?

Sean: Sure. So, you know, typically when people engage with us they start with an idea in mind. They have a product and they want certain types of functionality. And they want some AI inference capability. And they want to do it outside the data center. They don’t wanna have a network connectivity. They don’t wanna have large power. So, they want to do something.

So, I can just talk about some of the customers and things that we talked to. We’re seeing interest for things in hearing aids such as denoising and clarification. We see things from customers and interest in industrial for IoT or vibration analysis to say, “Hey, here’s the signature for preventive maintenance.”

We’re seeing interest for wearables such as I’m wearing on my wrist right now. We see interest in automobiles and various parts of the automobile. So, we’re seeing a whole variety of things that people are interested in across a lot of verticals.

Marc: So, just as an example the hearing aid?

Sean: Yes.

Marc: Akida in that use case would basically filter out what is voice or what is noise and skip the noise and just bring out the voice of whoever you’re talking to?

Sean: That’s correct. At an incredibly low power, very efficient way, low number area in the piece of silicon.

Marc: And so, if you go back in your announcements one of them said, “We’ve announced the second generation but the customers’ prospects are waiting for the actual platform to be available.” So, in terms of engagement, in terms of attraction, what are you seeing now that is different from a couple of months ago?

Sean: If I look across our pipeline of all the engagements, about more than half…much more than half, probably three-quarters of it is all around Generation 2. So that tells me it’s something that people really, really wanted to have. So, the interest level is incredibly high in Generation 2. If you look at the analyst reports that we got when announced Generation 2, overwhelmingly favorable and very strong position in the market.

Marc: And how far along in your sales funnel would you say some of these prospects are? Because it’s one thing to be interested and a second to look at it and start testing with it, but at the end of the day, of course, you want revenues out of that. So, can you shed some light on the sales funnel for some of these groups?

Sean: Sure. So, these are very serious engagements. These are engagements for customers that have timelines to create product. So, they’re not just saying, “Hey, show me this, and let’s figure out if there’s something to do.” Many of these customers have real projects to say, “I have a date somewhere out in the future to get something done with a timeline. We’re gonna create an evaluation. We’ve got budget to do that.” So, they’re very serious engagements.

And then when we engage, it takes a fair amount of time, typically it starts with a customer sharing their models with us. We’ll run some models and share the performance data with them and say, “Here’s what you can expect from us.” And so we do it that way. In addition, then we might turn over a board or a system and let them run themselves.

So, they take a fair amount of time, but we’re deep in several, several engagements. Again, what I wanna point out is, these aren’t just explorations, they’re projects where people have defined budgets and defined timelines.

Marc: And so, you don’t do that unless you’re very serious about actually implementing that.

Sean: Yeah, because it’s going to take a fair amount of resources on their half, too, right? They’re gonna have to dedicate some technical resources to do these evals.

Marc: And if you could sort of segment the interest from all these groups into industries, which one would you say, which industry would you say is sort of dominant among your prospects?

Sean: I think it’s relatively distributed. And it’s along the ways I’ve just talked about, a fair amount in wearables, a fair amount in hearing aid, a fair amount in industrial, and a fair amount in automobiles.

Marc: In one of the initial deals, I think it was MegaChips that was probably two years ago, almost, that was one of the big ones in terms of the first license fees coming in, right?

Sean: Mm-hmm.

Marc: How has that progressed the MegaChips collaboration? And as a classical example of a license fee and royalty model, how far along is that one in terms of mass production, which means which would mean royalties for you guys?

Sean: Yeah, so, we don’t talk about the customer themself in their progress, right? You know, it’s their business. Now, we work with MegaChips. We are in constant contact with MegaChips. We support MegaChips. But I really can’t comment on where they are in their revenue streams right now.

Marc: But it’s fair to say that sort of in general, when you’re looking at companies like this, they pay you a license fee, maybe more at a high level, and what sort of delay is there between getting that first revenue in from licenses and actual sort of royalties from mass production?

Sean: It certainly takes time. So, when somebody is gonna build a custom chip, it takes a fair amount of time. And typically, IP selection is pretty early in as part of that process. So, once they decide to buy their IP, then they’re gonna have to go ahead and complete the design of their chip.

And that takes anywhere from 6 to 12 months. And at that point, then, of course, when the chip comes back, they’re gonna do some tests, QA, and then eventually it goes into their end product, and then it goes out to market. So, it could be well over a year, two years at times.

Marc: If we change focus a little bit to the balance sheet. So, I noticed you’ve got about 17 million in cash. You’re burning roughly at four quarter. Can you remind us of the deal or the structure of the deal with LDA Capital for your funding purposes?

Sean: Yeah, sure. But before I do, I wanna add one more thing if I could to the last comment. Because, you know, I’ve got a lot of questions on this trip that I’m on right now about a recent announcement we did on a commercial model. We announced something with a system in our grader called VVDN. And in that, VVDN is going to create an Edge box. An Edge box is gonna be very small in its form factor. It’s gonna include our chip, the Akida 1000 in that chip, in that box. The box will be used for all kinds of use cases.

Some of it could be for security, some could be for retail monitoring, and things like that. It’ll be sold predominant by VVDN. We will sell it as well. We’re not gonna sell in volume with our salespeople. We’re gonna put it up on our website for people who wanna buy it. So, people have asked me repeatedly, “Is that a change in our business model to a chip versus IP?”

The answer is no. We’re doing that for a couple of reasons. We really wanna just prove out more workloads into the market, right? And there’s been an ask from system integrators to do this. So, we’re gonna do this with VVDN, and we might do it with others right now. But it’s not a shift in our business model. So, I wanna be clear about that. We’re just gonna try this and get some workloads out there. So, you’ll see some revenue, but we’re not expecting to be big revenue off the hardware side.

Around LDA, your question about, “How does it really work?” It’s a very favorable and friendly way for us to do our fundraising. What we’ve done, we provide a certain number of shares to LDA. We specify a period of time that they can really start to market. We set a minimum price that they can’t sell below. And they take 8.5% for that service. All the details are in our financial filings if people wanna know more.

Marc: And so, going forward, you’ll still be using that structure or other sources of funding that you’re looking at?

Sean: You know, as a CEO and our CFO and our board, we’re evaluating all options for funding, and needs, quite frankly.

Narc: Sorry, and…

Sean: And the need, right?

Marc: And the need. Yeah, of course. Yeah. All right. Last question. What can investors look forward to in terms of sort of milestones in the next 12 months?

Sean: I think the next 12 months are critical for BrainChip, right, in a lot of ways. Now, when I came here two years ago, I think it was roughly two years this week, actually, the company was just really exiting its R&D phase. Now, we had just released a chip, but there was, you know, minimal sales, minimal salespeople.

We had minimal marketing presence. We had just realized, you know, our Generation 1 was narrow and focused. So, this last period of time, a lot of work went to creating Generation 2, which we believe is the product that the company is gonna be built on for the coming years. We revamped our entire sales organization. We now have sales presence around the world. We have people in Korea, Japan, Israel, Europe, United States.

We’ve got a world-class marketing organization. So, we’re in that position now, where we’re now we talked a little bit about the pipeline as well. We’re in a very different position than two years ago. So, the next 12 months is critical. So, what can people expect from us? Certainly more partnership agreements.

That’s a critical part of our success. I made a comment at the AGM that we’re gonna accelerate our product development. So, look for more product announcements from us. And also look for announcements around engagement so we’re hopefully gonna be winning.

Marc: Exciting times then.

Sean: Very exciting times there.

Marc: Good stuff. Thank you very much, Sean.

Sean: Thank you, Marc.