BrainChip (ASX:BRN): Interview with CEO Peter van der Made

We spoke with BrainChip CEO Peter van der Made about the company’s commercialisation of the Akida Spiking Neural Network chip and the market opportunities ahead.

See a full transcription below.

 

Check out some of our more recent articles on BrainChip:

• Why ASX chip stocks have been on fire lately
• BrainChip (ASX:BRN): Welcome to the revolution!
• BrainChip (ASX:BRN): Why we like this stock so much

 

Stocks Down Under … Must-have information for the serious investor!

 

Get Marc & Stuart’s Top Picks

Improve your investment skills with our Investor Webinars

Learn about ASX-listed stocks with our 4 Stocks Down Under editions per week across all sectors

 

Get access to everything with a 30-day FREE TRIAL.

No credit card required!

GET A 30-DAY FREE TRIAL

 

 

Transcription

 

Marc: Hi, and welcome to “Pitt Street Research.” We are doing an interview today with Peter van der Made…or in Dutch, Peter van der Made…Dutch heritage there.

Peter: Yes, that’s right.

Marc: He’s the CEO of BrainChip, listed on the ASX. Welcome, Peter.

Peter: Thank you.

Marc: For people that don’t know BrainChip that well, in a nutshell, can you explain what you guys have invented/developed, and where you are now with the product?

Peter: Yeah, we started in 2004 with this project. The first 10 years was a lot of learning and developing and trying things out. In 2013, we moved to the United States, and we started going in fast forward. What we have developed is a different way of processing sensory information. So, what we’re getting in is sensory information from a camera or from another type of sensor, and we process the information in a very different way from this traditional way, where you have a microprocessor that goes around a little circle, and tries to interpret what it is receiving. Our processor is working like the human brain. The human brain is also receiving information through its eyes and ears and nose, etc., and that information in the brain is processed in the form of spikes, in a very…highly parallel way, so each cell processes information in parallel to every other cell. And that’s exactly what Akida does, it processes everything in parallel.

Marc: Right. So if you look at Akida, what problem does Akida solve compared to other products on the market?

Peter: Yeah, Akida really has five advantages over any other product, the first thing is it runs with a very, very low power consumption. That means that you can run very complex processes off a small battery. Like for an example, order declassification, it can run off a small one and a half volts AA battery for five months.

Marc: Right.

Peter: So, that’s one of the biggest advantages of the Akida technology. It’s also very fast, compared to other products that are simply using a DSP or something like that, that is still processing in the traditional way of going around this little loop, instead of doing everything in parallel. Our process can run at low clock speeds, and because it runs at low clock speeds, it is very efficient and very fast.

Marc: Right.

Peter: It also is learning on-chip, instant learning, just like your brain learns. You don’t have to have 1,000 images of a dog to recognize a dog, we can do it with a single shot. We show it one image, and it immediately learns what is the animal.

Marc: So, the chip gets better as it’s out there in the world, learning by itself? It gets better over time…

Peter: Yeah, either that, or you can train it up before you send it out into the world.

Marc: Yeah. But even after that, it would still get better as it’s in the field.

Peter: It could. It could, yes, if you wanted.

Marc: Yeah, right. Okay.

Peter: But also, you could use this facility for configuration in the field. If you have a car, for instance, that recognizes the driver, you don’t want to have to train it on the image of a driver. What you do is you have a driver sitting in the seat, and train it in a single shot, and now, the car recognizes the driver.

Marc: Right.

Peter: So then the car can adjust the seat and the radio station and the environment to the driver’s preferences.

Marc: Right.

Peter: One of the most important things is also, that chip is compatible with TensorFlow. And TensorFlow is an environment that is familiar to every data scientist in the world, so people can start using this chip straightaway. The chip is very small, very light, and has on-chip convolution, which means that existing networks that people have already developed over time, CNNs can run on this spiking chip.

Marc: Right. And I think one of the key advantages as well is that it doesn’t need a connection to the cloud, to the internet, to perform its functions, right? So let’s say an autonomous vehicle, it’s got one of these chips in it, as an example, there’s no time really to connect to the internet to interpret, you know, certain things and to come back with a response, it does it on-chip, so you know, instantaneously almost, you get a response from the chip.

Peter: That’s correct, yes. You don’t need to send anything up to the cloud, everything is processed right there on the chip.

Marc: Yeah. So, that’s a big thing in edge computing. Yeah, all right. And so, how big is the market, do you think, currently, for what you’re doing?

Peter: We looked at some forecasts from Tractica. Tractica is predicting that the market will grow very rapidly, the market grows to $60 billion size in 2025. So, we are looking forward to getting a significant part of that market.

Marc: Right. Okay. And in terms of revenues, so can you talk a little bit about that? So in typical semiconductor models, you’ve got IP licenses, you’ve got royalties, you’ve got some of this stuff upfront, the NRE, non-recurring engineering revenues, although that’s not what you’re doing it for, but it’s part of sort of the initial stage of commercialization. Can you talk a little bit about how you make money, and what sort of the near-term sort of milestones are?

Peter: Yeah, we have four paths of revenue. The first path, as you mentioned, is the sale of IP, and IP is then followed after the design of the product by royalties. The next path of revenue is to sell chips, completed chips. That path will kick in after we ship some modules. I expect that modules, development systems will take off first, for the simple reason that you plug one of these modules in, and it works right away.

Marc: Right.

Peter: So, we’re thinking of a cheap module, something that plugs onto a Raspberry Pi computer, that we can ship in large quantities, saturate the market with this, which creates channels for future IP and chip sales.

Marc: Right. Okay. And so, in terms of the first commercial products, so the Akida 1000, can you talk a little bit about where you are with that one right now?

Peter: Well, the Akida 1000, as you probably know, we had engineering samples earlier this year. Those engineering samples have been tested, found to work satisfactory. We then moved into creating a production mask, which has been completed and has been sent to TSMC. TSMC is now in the process of producing our chips, our first lot of chips, our commercial product.

Marc: Okay. All right. And so, those will be sold as commercial products, or still as sort of, you know, samples, or you know, testing products for potential customers?

Peter: These chips will go primarily into modules, we will ship those modules as commercial products.

Marc: Right.

Peter: There will not be huge numbers, initially, we will ship these first to a number of small engineering firms that are currently waiting for product.

Marc: Right. Okay. So, and then, you know, what people always look at when it comes to technology companies is sort of, you know, what’s beyond the horizon? Can you talk a little bit about your development roadmap for…I wouldn’t say spin-offs, but different versions of Akida, going forward?

Peter: Yeah. Akida is the beginning of a whole family, the Akida 1000 was designed to cover a very large area of the market. The first spin-off will be a smaller Akida, a smaller Akida in the sense that it has fewer circuits on it, and therefore it will be a lot cheaper. If you look at a wafer, for instance, that we have on the table here, the larger a chip is, the more area it takes on the wafer, the more area that it uses on the wafer, the more it costs us to manufacture. So Akida 500 we be very small, and is aimed at the market for refrigerators, home appliances, that sort of thing. So you only need maybe a few thousand neurons, rather than a million.

Marc: Yeah, okay. And look, this is a 200 ml wafer, you’re producing…you’re manufacturing on 300 millimeter, right?

Marc: Yeah.

Marc: So, there will be substantially more, about 30% more, typically, chips, on a bigger wafer, right? So, that’s where you get your economies of scale?

Peter: That’s right.

Marc: The bigger you go, the better it is. Yeah.

Peter: Yeah. Yeah.

Marc: All right. So lastly, and this is a question that we get back from our readers, not specifically for BrainChip, but in general about the semiconductor market, and how tight production capacity is, really. So you know, the likes of TSMC, the Foundry’s of this world are, you know, completely sold out of capacity, probably for the next two years or so, and some lead times of simple, basic modules can run up to 12, 18 months right now. So, how have you been able to secure your sort of production slots with, you know, TSMC, and how do you see that going forward?

Peter: We have an excellent relationship with Socionext in Japan, Socionext has already booked production slots with TSMC. So, we have been assured that there is no issue with shortages for the Akida chip.

Marc: Right. Okay, that’s good news. That’s good to hear. All right, thanks, Peter.

Peter: Thank you.