2021 BRN Discussion, page-31225

Transcription (Stockdownunder interview with PVDM)

Marc: All right. So the last presentation is from Peter van der Made. If I could invite Peter up to the stage, please. Peter is a fellow Dutchman I’m glad to say. So there’s a lot of semiconductor expertise actually in the Netherlands. Historically, he’s come from Philips Electronics and from their Philips Semiconductors and ASML so there’s a lot of expertise there. And Peter, good morning to you.

Peter: Good morning, Marc. How are you doing?

Marc: Pretty good. Pretty good. Thank you for joining us today. I was just talking about our background in the Netherlands in the semiconductor space. I suspect your background is potentially from Philips Semiconductors as well. But like I said, there’s a very big ecosystem and infrastructure in the Netherlands around this. A lot has happened for BrainChip recently including a very nice announcement last week. So without further ado, I’ll hand the floor over to you.

Peter: Yes. Thank you, Marc. It’s really good to have the opportunity to tell you about all the exciting things that are going on at BrainChip. I’m not sure if you can see our slides at the moment. Is that on, Marc? There they are. Yes. So let’s go to the next slide. Where are we? There was a button somewhere to flip. There they are.

So it is our statement, our disclaimer that says that we did our best to bring the best possible statements about our company, but that some of the information is available on the external sources that we cannot verify. You can read it at your leisure when you get our slides. So if we could start with the BrainChip mission statement. BrainChip was started back in 2004 with the objective to build safer, more sophisticated artificial intelligence processes. That’s exactly what we have done with the introduction of our Akida 1000. So with the Akida 1000, we have accomplished our first step in this way, a commercial neural multiprocessor that is unique in the markets. So this component has been recently introduced. We are now building boards with this component and shipping those boards to our early access customers.

The component is very unique because it’s very low power consumption, extremely low power consumption, which is good for the planet. Because if you look at current artificial intelligence in the computing way, where you’re completing the outcome, it generates enormous amounts of greenhouse gases. So if you look at the collected data centers of the world, it’s something around 600 megatonnes of greenhouse gases. And Akida 1000, if you look at a single instance, Akida 1000 is estimated at about 97% to 99% more energy-efficient in cloud processing. Or even if you look up GPUs, they are very power-hungry, up to a thousand times more power than Akida actually is. So we have also looked at healthcare. We believe that we can make a big difference in the healthcare field where our sensors pick up for instance [inaudible 00:05:04] blood, VOCs in breath and other medical samples. [inaudible 00:05:10] classified as faster and more accurately than any other method that is currently around. And even learn on the chip.

We also looked at safer workplaces where Akida can eliminate the risk in the workplace by monitoring, classifying air quality, automated quality control, visual classification, and tactile sensing. MetaTF is our development platform. MetaTF plugs into TensorFlow. It’s easy to use. Anybody who knows how to use TensorFlow will be able to use MetaTF. It’s self-contained, it’s fast, [inaudible 00:05:51] effort response and API that’s easy to use. So it provides for the onset learning and training, and a seamless environment for any developer that understands TensorFlow.

So a little bit about the background of BrainChip and what we are doing at the moment. At the moment we have development centers in two places in the world. We are in Perth in Australia with the Research Institute. We have a software development and testing center in Toulouse, in France. We have chip design and sales and engineering in Southern California. And our corporate headquarters in Sydney. We do the assembly and layout of the chip in Japan. So the current situation of the BrainChip share price on the 25th of November was 63 cents with a market cap of just over a billion dollars. We have 1.8 billion shares outstanding, with 1.3 billion in float. And as management hold just under 20% of the company. We are growing rapidly. We are employing people at the moment, especially in sales and marketing, and looking at about 70 people with our, as I mentioned, headquarters in Sydney.

This is our management and board. Between the five people in the management team, we have something like 180 years of experience. Rob, he’s our sales and marketing guy. He is ex ARM. Ken, CFO. And he is ex Virgin Galactic. So really going reaching out into space. Anil is our chief design and chief development officer. Anil must have made at least a hundred chips in his life. You know a little bit about my background. This is my third company I’ve started. The first company, PolyGraphics, high resolutions graphics back in the 1980s, early ’80s. I then went on to teaching at university for a while. And then after that, I created the new company called vCIS, which was later purchased by IISS and later by IBM. All the patents of the company are now in the hands of IBM. And Sean, who just joined us, many years with HP, and other large organizations. And so the seasoned executive in sales and marketing and building large companies out of early beginnings, especially in sales and marketing. And that’s why we focus on Sean and look at candidates for CEO.

Our board of directors, I’m also part of the board of directors. Many, many years in the semiconductor industry, in the finance area. Geoff, ex-Commonwealth Bank. And Antonio with many years in the industry as well. So we have a very seasoned team. Also, the people who are working for us are extremely well experienced. Oops [inaudible 00:09:32]. If you’re looking at our financial position at present for a burn rate, BrainChip has sufficient to actually run for two years. Of course, our burn rate is going to change over these years. We’re going to employ more people. But at the same time, we are counting on an explosion in sales. We’re looking forward to an explosion in sales on that Akida 1000 chip and its modules, and also the IP. We just signed up, as you may have seen, MegaChips, which is a large organization that makes chips for many different applications.

We also finalized the American depository receipts that makes it available for U.S. institutional investors. And we have upgraded our listing in the United States to the 2X level, which means that institutional investors can now invest in BrainChip in the United States. The company ended the previous quarter, as we have published in our listing, almost 24 million was in cash or 32 million Australian dollars. That’s U.S. dollars.

The patent’s protection is very important. As I said, our technology is very unique. It’s been a 16-year development path. So we want to make sure that this technology is protected. And our first patent stems from 2008. We filed that same patent in 2007 in Australia. And the patent has been followed up by new patents. We actually have a number of patent engineers within the company now who are pursuing these patents because in the last couple of months a lot of the patents that had been sitting around for a while had been granted. We have two patents that are allowed, patent [inaudible 00:11:47] seven and nine, which will soon be granted. We expect that it’ll be granted within the next month.

We are building out our patent portfolio to go worldwide as we are marketing the chip and the technology IP worldwide. So therefore we have filed in all these different countries. And in the next year or so, we will see at least hundreds of these patents being filed in different countries, and subsequently sort of granted.

So if you’re looking at our products, the Akida 1000 is the major product, which is the chip. The neural multiprocessor mimics…and I’ll look it up in the dictionary, it mimics neurobiological architecture present in the neural nervous system. So it doesn’t work like a processor that is applied in your laptop, it works like a mini-brain. It is very good at recognizing things, recognizing smells, recognizing images, recognizing tastes. We have done all these things as examples and made them available to the market. We also have an artificial retina image to spike converter. What the eye is doing, it takes images, converts those images into spikes that then the brain is processing. We do something very similar with our artificial retina image to spike converter.

The chip is up to 10 times more energy-efficient than any other edge AI processor. Except if you’re looking at very tiny things that can’t be really used for anything. But anything that isn’t the same class as the BrainChip Akida processor is using at least 10 times more energy here. And it’s up to a thousand times more energy-efficient than GPUs.

We also priced the chip for use in hand-held solar-powered equipment. So that at the [inaudible 00:14:01] at the dollar store for a chip, so we priced this chip at a very competitive level, somewhere between $15 and $25 in quantities. Other point is real-time learning. We do real-time learning, which is unique in the market. We have a video for instance where we show little elephants to the camera, and Akida classifies them, you can tell if it’s an elephant. We can then show it a picture of an elephant in the wild and it will recognize the elephant from any angle.

This is really unique in the market. Nobody else has been able to do this in a single shot. It learns very much like a child learns. Take a child and tell them it’s an elephant, and for the rest of its life, the child will know what an elephant looks like. You don’t have to show it a thousand examples like you do with [inaudible 00:14:54] deep learning.

The product we’re building around Akida, of course, we have MetaTF as a development environment. We also have USB stick. We have the M.2 Class insert board and the PCIe board for development as a development kit. This includes free sample codes with a development suite for visual, sounds, all those vibrations at that top classification, although very high accuracy.

So in comparison to the rest of the industry, if we’re looking at traditional AI versus neuromorphic AI, if our consumption is very high of traditional AI, neuromorphic AI is very, very low. Learning in traditional AI is expensive and time-consuming. It takes weeks to train a network. We can do once real-time learning, which takes milliseconds. Traditional AI needs cloud connectivity, Siri, for instance, and Alexa. They send all the information that they receive up to the cloud. It’s particularly a problem when you’re sending images, they can be hacked. In Akida’s case, neuromorphic AI is completely independent of the cloud. All the information is processed on the chip itself, and the classification is produced by the chip itself. So it’s very secure. The data that stays within the chip cannot be hacked.

Also, if you have Siri or Alexa or any other cloud-based AI nest, for instance, it requires an internet connection. Once it hasn’t got an internet connection, it doesn’t work anymore. And because of the need to facilitate information up to the cloud and then get it back as an answer, the latency can be quite high.

So here we’re looking at the 10 problems that Akida 1000 is solving in the industry. The IoT bandwidth is a problem that it’s building. If you have 40 billion as a forecast, 40 billion new units that connect to the internet, IoT devices. IoT bandwidth is becoming a serious internet [inaudible 00:17:25] becomes a serious problem with increased latency.

Power consumption, if you’re looking at CO2 emissions, Akida 1000 is extremely energy-efficient. Availability. If you’re in the middle of Nullarbor, you won’t have any internet connection, for instance, with Akida, you don’t need an internet connection to keep the unit working. So you don’t get the message that it needs an internet connection.

Training. Akida can be trained very rapidly with one-shot on-chip learning. Personalization, which is a very important thing, for instance, in the car industry, somebody gets into the camera and Akida can classify who that person is. Then learn instantly who that person is if it’s a new owner of the car for instance.

Portability. It’s small, it’s very light. Excuse me. It’s very small, it’s very light. And it allows AI in portable equipment because it doesn’t get hot. It uses no power almost. Very little power. So it doesn’t get hot, it doesn’t need any cooling. It can be used as a standalone module. It has an own chip processor. The on-chip processor, its only function is to preprocess data and receive data. The neuron fabric [inaudible 00:18:46] all the neural network work. So it’s very low cost. They are engineered for low-cost applications. It’s modular. You can modify the IP very easily to fit the manufacturer’s requirements. And IP licenses are available. And it’s an easy development environment that is familiar to many different data engineers.

So here we’re talking about the bandwidth problem [inaudible 00:19:24] devices forecast to be completed for bandwidth. Of course, Akida doesn’t need any bandwidth. We solve that problem. Data centers are growing with 20% per annum. And by 2030, if we continue going the way we’re going today, data centers will consume 30% of global electricity by 2030. So Akida can solve that problem because we got not processing [inaudible 00:19:53]. We get processing distributed over many different locations. And those locations can be powered by battery or solar power.

So talking about here is the runaway power of AI processing. At the moment, we only have the figure of 2016, which was 416 terawatts. That is about 30% more than the whole United Kingdom. Since 2016, it has been increasing at about 15% to 20%. So today’s estimation of CO2 emissions is something like 600 MegaTons. With the distributed processing where you have Akida in every device, you would be far more energy-efficient and avoid this problem.

So the Akida processor it has a large number of interfaces on it. It can talk to normal computer equipments. It has these M-Plus CPU on boards for data processing. It has the video frame interface. The artificial retina. I talked about information from the video camera can process into events or spikes, which are then processed on the neuron fabric and you get an output that classifies what is in-frame. We also have an external memory interface. We have multi-chip extensions, so you can put 64 Akidas on a single board. And I already mentioned the on-chip one-shot learning.

So it’s a very sophisticated device. It’s very exciting that we have this device ready at the right time for the right market. This is the forecast by Tractica. The Tractica forecast between 2018 and 2025. We’re looking at the 2022 bar here that 450 million forecast, 450 million A SIC that all could use Akida IP. With 700 million accelerated chips that could also be the Akida chip, the Akida 1000. And you see that they’re increasing very rapidly in the years to come. And this is a very exciting slide for us because this bottom part of the graph is where we operate.

The comparison of Akida through all the other components that are out there. So we have a very well documented competitive analysis, and this graph comes out of that. In place of Akida 1000, it ticks all the boxes. It’s micro to min power use. It’s real-time on-chip learning. It’s TensorFlow compatible. You can use it in standalone mode. The on-chip convolution, which means that you can look at images [inaudible 00:23:11] networks. It’s available as IP and it’s a very green technology.

If you’re looking at IBM TrueNorth, which is an older chip, it’s also a test chip. It’s not a commercial device like Akida 1000. It ticks the first box but none of the other ones. It’s a green technology because it has also low power, not as low as Akida.

Intel Loihi. Intel Loihi as well, it ticks the first box, it’s minimal power use. If you want to do learning, you have to program. It’s got its own environment called LAVA for learning. They have on-chip, now 686 processors, which cannot be very power efficient. They do say that they do convolution but I think it’s being done by the x86 hardware. And it’s not available to IP.

Then there is what I call deep learning accelerators like the Coral TPU and the NVIDIA Jetson. Those things are not very low power that through the [inaudible 00:24:25] Watts for the TPU, [inaudible 00:24:27] the CPU next to that. VLAs are in the range of 5 to 10 Watts. They are Math chips, they’re not doing any neural networks. They just perform multiplication. They are TensorFlow compatible but they do not check any of the other boxes. They do not do on-chip convolution. They have to do convolution on your CPU. They are not available as IP and they’re green because they use about the same amount of power for instance as a data center.

So these are our websites, and basically, you can find BrainChip on LinkedIn. Our videos [inaudible 00:25:14] at our videos. And I think that we are, potentially, Marc…

Marc: Excellent. Thank you, Peter. So we’ve got a number of questions coming in, some of them center around your current partnerships that you’ve announced previously, as well as the early access program. So let’s break that up in two. Can you talk a little bit about the feedback you’re getting from your early access program collaborators? And I know the question that we got out of this might be hard to answer, but do you have an idea of the percentage of partners or people that are trying this out that could convert to commercial deals?

Peter: The customers to convert to commercial deals? Of course, to take a little while for them to get their projects work out to the point where they are ready for a commercial deal. We’re particurlaly excited about the NASA people who have given us feedback. Very positive feedback about what they’re doing. Of course, they’re getting the latest generation of the chip. At the moment the production chip, which has been delivered to a number of people in the EIP Group. You cannot mention names, so EIP customers. Unfortunately, because they’re under a non-disclosure agreement, people do not want to splash what they’re working on in their labs out there in the market [inaudible 00:26:47] you know, what’s going to come next, not until they’re ready to do so.

Marc: Okay. Yeah. I understand that, especially in the semiconductor industry, it’s highly competitive and it’s always very sort of hush-hush. And it’s something that, unfortunately, the ASX doesn’t always understand. But is there anything in general that you can talk about with these partnerships that you have with NASA, Valeo, Magic Eye, Ford? Any sort of a common denominator that you see in these conversations with them?

Peter: Yeah. The common denominator would be that all of these companies are very excited about technology.

Marc: All right. That’s clear. A question from Dan about when we can start to see physical devices contain Akida, and I’m assuming this would be commercially. If you can comment a little bit on that.

Peter: Yes. As people are developing with Akida, they need to develop their boards, they need to develop their software, they need to prepare the whole market for these products. I’d say that in the next year we will see… I expect to see commercial products that are out there with Akida in it.

Marc: Okay. And in terms of revenue, some of the units via the recent batch of chips that have been completed, been distributed to partners. Is there anything you can say about what sort of revenue? Maybe not the level of revenue, but if there is something that, in the current quarter, that you will be reporting on soon. Because I seem to remember there was definitely a commercial or at least, you know, revenues coming from these chips, these initial chips.

Peter: Yes. And the revenue will be showing up in our 4C. I can’t really comment on that at this stage until we’re ready for these at our 4C at the end of this quarter.

Marc: Okay. Regarding the Akida 2000 and the timeline specifically for that one, can you talk a little bit about the development timeline?

Marc: Yes. Akida 2000 is very exciting too. We are developing that here in Perth. We have an excellent team here in Perth of highly qualified people who are working on Akida 2000 and Akida 3000. So we’re really thinking ahead here. Akida 2000 will be optimized for more complex network architectures, such as [inaudible 00:29:37] transformers. And for a very large part, those simulations of those chips are… At least Akida 2000 is already…the simulation is working already. And we are getting ready to hand it over to engineering. So Akida 3000 is going to a cortical network. Cortical networks are based on the way the human cortex works. They’re still a lot of open questions about, especially in neuroscience, how that exactly works. We’ll be building models of the cortex and see how we can apply that to a real commercial environment. Of course, everything we do, of course, even though the science is fascinating, we have to always keep in mind that we are building commercial products.

Marc: All right. One interesting question I saw earlier coming from Nathan about the advantage you have in terms of time versus competitors. Previously you stated two to three years advantage versus competitors. With the Akida 1000 now almost sort of commercially available, do you still think your timeline or the time advantage is still two to three years or has that changed?

Peter: Yes, we are working very hard to make sure that we maintain that advantage. So Akida 1000, we estimate to be two or three years ahead of the market. For instance, we have convolution and hardware not by a processor. We have real-time learning that nobody else has been able to manage. So we definitely have a two or three advantage in Akida 1000, plus we have commercializing that product. But to maintain that advantage, we are working on Akida 2000 and 3000. So we stay ahead of the markets. With Akida 3000, we’re probably about five or six years ahead at the moment.

Marc: All right. Well, last question, Peter, because the audio is still, apparently, not great. What we might do actually is just send you these questions so you can come back to us by email and then we can put them up on the website. But just the last question. With the new CEO, Sean, it’s very early days, but some questions come in for him as in how has he gone in the first sort of week, basically?

Peter: In the first week, yes, he’s extremely eager. He worked over at the… I got the questions about the business plan and technology and what we’re doing over the Thanksgiving weekend. So he’s extremely eager to get started. And yeah, we have a great working relationship. I really enjoy working with Sean.

Marc: Excellent. All right. Well, Peter, thank you very much for your time. We’ll send through those questions to you by email so we can get answers to those. And again, for everyone that’s here today, we’ll put up on our website.




https://stocksdownunder.com/video/brainchip-at-the-stocks-down-under-semiconductor-conference-on-1-november-2021/