Brainchip in the News, page-29

no apologies for reposting this announcement from less than 3 months ago.


Chief Executive Officer’s AddressBrainChip is entering an exciting new era as we move from an R&D focus to a manufacturing and sales focus. We have an exciting product, a product that is entering the market at the right time, with the right features that enables clients to design Artificial intelligence into products that are powered from a small battery, wearable electronics, or remote sensing equipment that does not require an internet connection to operate. Information is processed right next to the sensor. This is “the edge” – no internet connection to the cloud is required. Performance is not compromised by transmission delays or bandwidth problems, and security is enhanced because nothing needs to be transmitted. We are starting here with our standard disclaimer. We use forecast information about market size and applications from 3rd parties. Like weather forecasts, a lot of effort goes into getting these forecasts right, but they cannot be right all of the time.You may have seen this video already. It mentions ‘neuromorphic event-based processing’. Now that is a mouthful. What does that mean? Neuromorphic means that the chip is processing information in a similar way that the human brain is processing information. Our brains do not need a computer program to function. The brain learns everything it learns when it comes across images, sounds and other senses. The brain learns instantly. We do not need a thousand images of a dog to recognize a dog. A child will see a dog once, is told that it is a dog and then will recognize every dog.The circuits inside Akida have a similar function to the cells in the human brain. These circuits are processing incoming information in parallel, all at the same time. There is no program code to make them do this, they are designed to imitate the processing method of the cells in the brain, including the learning function. Neuromorphic literally means ‘like the brain.’Event-based means that it only processes things that are important. The brain does this too. Only when something changes is it necessary to process anything. For instance, if you have a camera pointing at a room that is empty the image is processed once. Then no further processing is needed until someone walks into that room. Then you can decide if they person is allowed to be there or not. This saves a lot of power.Page | 4 Artificial Intelligence is a fast-evolving market. Some of the applications we have come across so far are remote controllers, financial forecasting, cameras, food safety, food quality and taste control, intelligent refrigerators, disease detection in breath - and blood samples. We expect that there will be many more. Artificial intelligence is a technology trend that will see use in many everyday products. It is likely that manufacturers who neglect to incorporate AI into their products will fall behind and lose market share to others who have incorporated intelligence in their products.I will give you just a few examples of how beneficial AI will improve the quality of our lives. Refrigerators equipped with Akida could smell if anything is going off and tell you what it is likely to be, avoiding food poisoning. According to Google, each year, 48 million people – 1 in every 7 - in the US get sick from eating spoiled food. Of those, 128,000 are hospitalized and 3000 die. (search food poisoning statistics)The display on the fridge could give an overview of the food in the fridge, and how long it will stay fresh. Doorbells could recognize the person at the door and alert the occupants.Factory robots are programmed which is a slow and tedious process. In the future they will be equipped with Akida and they can learn their task quickly – a big cost saving. Self-driving cars are also autonomous machines and will benefit from continuous and incremental learning, something only Akida offers.Your car will be able to recognize faces, making for a more personalized experience, for instance by setting the radio to your favorite station. The same Akida chip can be used to see if the driver is paying attention. 20% of all fatal road accidents involve driver fatigue. Akida can also monitor bearing vibration and engine noise for safety and preventative maintenance.In medicine, Akida can ‘smell’ and classify a variety of diseases using the right sensors. We announced our cooperation with Nanose, who build a room-temperature sensor for the compounds that are present in exhaled breath. Using test data sets that were collected in clinical settings the Akida chip can detect Covid-19 with 93% accuracy, much better and faster than a PCR test. We also signed up Biotome here in Perth who are working on a diagnostic blood test that will incorporate Akida.Over the next 5 years we expect to see a proliferation of Akida into all kinds of beneficial AI applications that will positively improve all aspects of life.Brainchip offers several avenues to enable and support the explosion of AI products that we expect to see over the next 5 years. MetaTF is a software tool that works within the Google TensorFlow environment. MetaTF is the development tool that is common to all Brainchip products. We have chosen to use TensorFlow because most data scientists are already very familiar with it. There is no new programming language to learn, and data scientists can dive right in.Page | 5 The Brainchip Akida IP is offered as an encrypted box that is dropped into a manufacturers chip design. It has been configured by Brainchip engineers for the number of nodes that were licensed, depending on the available space, the power budget and the performance that is required. For instance; The odor classification network can run in just four nodes. The Akida IP has been designed to be modular and can be tailored to fit the requirements. An IP license is followed by royalties that are payable on each product that is manufactured.The Akida chip - AKD1000 – is offered as a low-cost device for development, prototyping and small series manufacturing. We designed this chip to fit a wide range of applications, with many interfaces, a microprocessor, and a neuromorphic processor array of 80 cores. Once a requirement goes over a million units, we recommend licensing our IP to build a custom chip. This would be the case for instance for consumer goods.Our design partners receive training and tools to design custom boards for Akida product clients.Brainchip has designed and will be manufacturing at least three different modules. The M.2 size module shown here will be used in laptops. Most laptops have a M.2 slot. The USB3 dongle can be used in any USB port. We also have a PCIe board that plugs into a desktop computer. Solution providers will use these modules to build products.An exciting new product is the Tachéon Pi module. This module plugs into a Raspberry Pi and this makes a complete embedded system that can be integrated into products straight away. It makes a great educational kit for people to familiarize themselves with Akida as well.MetaTF, together with our modules are the tools that quickly enable engineers to use Akida in a wide range of home appliances; intelligent washing machines that sense washing loads, temperature, and wastewater. Fridges that know when food is off, control systems and preventative maintenance in cars, intelligent doorbells that recognize people, driver assist systems and thousands of other applications in smart home and industrial sensing and safety.A trend that has been observed is to move intelligence to “the edge’. What is the edge? Basically, it is every piece of electronics that does not need to connect via the internet to a large data centre to work.Most of today’s intelligent ‘edge’ devices – like Amazon Alexa – send everything through the internet up to ‘the cloud’ – which means nothing but connecting to a huge computer system in a megawatt-burning, water-cooled data centre. The data-centre colossus does all the heavy ‘AI’ lifting, it processing the remote data and sends the answer back to Alexa. Here a second delay does not matter much, but if you are in a far moving car it could be deadly. The next evolution of this process was to set up regional data centres, closer to the user products for distributed processing. This solves some of the problems. It is faster because transmission lines are shorter. The regional data centre is still a large behemoth computer that burns many hundreds of kilowatts and needs forced cooling. The security threat is not diminished. Bandwidth and latency problems are still significant. Maybe now you travel 15 meters before your car gets a response, but that is still deadly.Page | 6 Edge of the cloud platforms bring processing even closer to the application devices. These are smaller computing devices that do some local processing and communicate with the cloud. Delays are still significant because these computing nodes are not as fast as the large data centre computers. Security is still a problem because anything that runs programs can be hacked.- With over 30 billion IoT devices and 10.3 billion computers connected to the internet over the next 4 years, there will be insufficient bandwidth. Metadata, the output from Akida, can solve this problem.- The delay to get information back will increase because of this, causing a bottleneck.- In a self-driving car at 110 km per hour you travel 30 meters per second. Much cango wrong in a second.- Devices that rely on the cloud do not work at all where there is no internet, or theinternet is down.- Connected devices, such as cameras, medical diagnostic applications, can behacked causing a serious security threat.Akida represents the advanced far edge – sensors and application devices that are far removed from any data centre and don’t need to have an internet connection at all to work. All processing is done on Akida, at the point of acquisition.The Akida neural fabric consists of simulated brain cells, and they do not run a program so the neural network cannot be hacked by inserting malicious code. Akida is fast, and delays are minimized. Security is improved, processor efficiency is improved, internet bandwidth is improved or eliminated, and Akida can continue to learn after leaving the factory.In parallel to developing Akida and the MetaTF development environment, we have endeavoured to develop an expanding range of application examples. We are not a solution provider – but a chip manufacturer.We have examples for Akida in real-time learning, odour classification that can be used for food safety and disease detection, bacteria in blood detection, 3D point cloud object recognition, air quality classification, cyber security and access control security, anomality and fault detection. This covers all the main areas that you see here.Expands to many different applications, resulting in expanding opportunities for the Akida neuromorphic processor chip.In healthcare for instance, we can sniff and detect disease or monitor blood samples, but also see use in robotic surgery systems, X-ray and ultrasound image interpretation, patient statistics, and patient monitoring systems (artificial nurse).Figure 1 gives a quick comparison of Akida to the biological brain and standard deep learning methods. Important things to notice here are the power consumption, microwatts – that is millionths of a watt to milliwatts. To put that into perspective, an odour classification system could run on a penlight battery for 5 months. Akida is a ‘green’ technology. It alsoPage | 7 learns in real time on the chip, and it is very efficient and fast compared to the ‘old’ GPU method.Our current examples cover all five senses of humans:• Visual object detection, manufacturing• Accelerometer & tactile sensor, bridge resonance fault detection, robotics• Auditory – speech and sound recognition, key word detection, engine noiseclassification, presentative maintenance• Olfactory – detection and classification of odours, detecting disease in breath,detecting spoiled food, food manufacturing QC, air quality• Gustation – food and drink taste controlAKIDA AKD1000 has five main advantages over a large data centre behemoth computer. Akida does the same job, but in a different way. In the same manner as the brain, it is processing data simultaneously in many individual simulations of brain cells. It uses the same learning method as the brain, which learns instantly. It is thousands of times more power efficient. It is small and light weight, which is an advantage for flying machines such as drones, aircraft or in space craft.It generates almost no heat because it uses very little power.And we have on-chip convolution. Convolution is a technique that is used in Deep Learning networks. This makes it possible for the Akida neuromorphic processor to run existing Deep Learning networks with extreme efficiently.The future looks bright indeed. With all the advantages that Akida is offering the expanding Artificial Intelligence industry we are happy to move from a R&D focus into production, manufacturing of modules and sales. We released the design of the Akida chip to our production partners Socionext in Japan. Akida is not a simple chip. It has many layers which are processed at one layer per day. We expect to have several thousand working chips by August this year. We received and tested engineering samples in 2020, many of which have now been delivered to Early Access Partners.Our sales and marketing team are gearing up to hit the road running. Our priority is the enable design partners and solution providers with the tools they need to start integrating the Akida technology, either as a chip, a module or as IP.The BrainChip Sales team will continue to drive revenue through licensing of the IP, chip sales and module sales, as well as expanding our design partner program.Earlier this year we received our first order for an IP license. That project is running well, and the client is happy with our level of support. We are also working on a Proof of Concept with a large European manufacturer of car electronics systems. We have accomplished much with a small team, more than some of the large guys with deep pockets and large teams have.Page | 8 The 7 stages in the sales process:Creating awareness, through online blogs, trade shows, direct sales, social media, scientific papers and articlesConsideration: does the product match the customers’ requirements. We focus on sensor data classification, wearable electronics, medical diagnostics, and consumer electronics, and that includes all the many categories I showed earlier.Evaluation: The customer engages and evaluates the Akida technology in their application Customer Support: during the evaluation process, Brainchip engineers and scientists provide the client with support to design the best possible solution to integrate the Akida technology.IP licensing: Once the evaluation process is finished the client either purchases an IP license or places an order for AKD1000 devices.Development and testing: The clients develop their own products and tests the products functionally and market acceptance.Production and sales: At this point the client buys quantities of Akida chips or in the case of an IP license starts to pay royalties on each product manufactured.This table gives a quick overview of the type of devices and functions enabled by the Akida processor.Every device can be a wearable electronics device, can be configured by the user, can be battery operated, and can be used in remote, or very remote sensing – AKIDA checks all the boxes.In autonomous machines, Akida enables user configuration, solar power or battery power operated, and remote sensing. In augmented reality applications, Akida enables Artificial Intelligence in wearable electronics, user configuration (real-time learning), and battery operation. Artificial Intelligence in home appliances, refrigerators, washing machines, TVs, microwave ovens and entertainment systems to name a few, Akida enables user configuration, and intelligent monitoring or sensing. It could put on your favourite channel when it sees you sitting on the couch. In all appliances, it increases security and privacy because nothing needs to be uploaded to the internet. All sensory perception is processed locally, on the AKD1000 chip.This slide shows the unfair advantage that we have over competing products. Only the Akida AKD1000 checks all the boxes.We have Micro- to Milliwatt power consumption, real-time on-chip learning and training, a familiar TensorFlow development environment that a data scientist already knows, the possibility to run in stand-alone mode, and we have on-chip convolution to make it easy to port existing Deep Learning networks to the Akida AKD1000.IBM have a neuromorphic chip called TrueNorth. First, this is an experimental chip. It is very large, about 2.5 square cm and therefore very expensive to make. It does not learn, does notPage | 9 have on-chip convolution and to use it you must learn a new language called the Corelet language.Intel have also developed an experimental neuromorphic chip called Loihi. It has a programmable core if engineers want to add their own learning algorithm. It does not have on-chip convolution, and to use it you need to learn the complex neural engineering platform. Both these chips are low power, but not as low as the Akida AKD1000.Deep Learning Accelerators – or DLAs are math chips, like a small version of a Data Centre. The neural network is executed on the computer but multiplication and additions are accelerated on the DLA chip. They do not support any neural operations, no learning, cannot operate independently from a computer and generally use 2 to 5 watts, or 10x to 1000x more power than Akida, depending on the application.This is 3rd party forecast of the market for edge devices up to 2025, prepared by Tractica Research. Many of those devices could be using the Akida AKD1000. You can see that by 2025 the market size reaches nearly $58 billion distributed over industrial uses, mobile uses, consumer goods, automotive systems (ADAS, preventative and safety systems), and drones. I expect that air and space craft will also be included in this group.The prospects for the Brainchip Akida technology are huge. Many, if not most of these systems will need an Artificial Intelligence processor that is small, light, fast and power efficient. We deliver all those capabilities in the AKD1000.The number of ‘Internet of Things” devices will triple between today and 2025. All these devices will be competing for internet bandwidth. IoT is a category of machines and this is different from the ‘Internet of People’.With on-device near-sensor processing, enabled by Akida, internet bandwidth problems can be solved, even when metadata – that is processed already by Akida is uploaded to a central system. For instance - A video stream takes a lot more bandwidth than a simple text warning that an unauthorized person has entered the room.We expect that many of today’s ‘dumb’ devices will be replaced during over this period by intelligent, Akida based devices. In addition, there will be some 20 billion new devices. This is another great opportunity for the BrainChip Akida technology. You can tell we are excited about those prospects.If we split up that 58 billion dolar market into market segments we get the chart that is shown here. All of these categories match up with the ‘expanding opportunities’ slide that I showed earlier in the presentation. The Akida technology is flexible and can be used in a large number of very different application fields.The IP licensing model that we use is financially attractive. After the initial licensing fees in successive years, the client pays royalties on each product that is manufactured. Royalties continue over the life of the product. This has a cumulative effect – royalties stack up overPage | 10 the years to come. Every new license has the potential to add a new layer of income to the stack of royalty income. We are right at the beginning of this evolution, with our first license project progressing well.The AKD1000 chip is in production and we expect sales to take off soon after. Initially, we expect to sell complete modules while our clients are developing their own products and boards. Those modules will contain the AKD1000 chip and the MetaTF tools.This is where we expect the first production chips to be absorbed. Sales are expected to increase once clients have finished their development work and are integrating the AKD1000 into their products. Module sales are likely to continue into the future, followed by new products such as the AKD2000, which will be a chip with additional features to execute sequence learning networks, known in the industry as LSTM and Transformer networks. Important parts of the AKD2000 are already working now in the lab here in Perth. The AKD3000 chip is in development and will be aimed at capsule networks that are under development at Google and the cortical neural networks of the future.I already gave away the clue to this slide; Akida AKD1000 is not the end, but the beginning of an exciting range of advanced neural network products, each aimed at a specific market segment of this evolving technology.I’ll illustrate the reason for these products in an example:Today, if a plastic bag blows across the street, a car equipped with AI sees an object and hits the brakes, or worse, takes evasive action.Our future networks will be able to learn the difference between a plastic bag, blowing across the street and a rock by their behaviour. They will learn from sequences of events and from observing behaviour. A rock does not get blown in the wind. A door has a handle that makes it open. Objects have not only a shape, but also behaviour and a location in space. Our brain understands and predicts the behaviour of objects and sounds. We aim to build that intelligence into future products so that we enable our clients not only to build intelligent products, but safe and beneficial products.The decision to end Mr. DiNardo’s employment was mutually agreed upon between the Company and Mr. DiNardo. BrainChip is moving from an R&D pre-production phase into a commercial production phase.It was agreed by both parties that the Company needed to recruit a CEO with the appropriate skill set, the drive and vision to lead the business through the upcoming commercialization phase, and beyond. Lou joined the company in 2016 when we were still in a research stage. Since then we have made much progress, cumulating in the AKD1000 chip being in production, an upgrade to the OTC QX in the US, and the ASX-300 in Australia.Page | 11 To attract the right people, we must be attractive. Part of that attractiveness is to issue options. Brainchip has to compete with the big boys, Qualcomm, Intel, IBM, and a dozen others who can offer incentives and salaries that are beyond us. To attract the right quality of people we grant options as a long term incentive plan (LTIP).Each option has a strike price that is equivalent to the share price at the time that the employee joined. They can buy shares at that price when the options vest. It is therefore in their interest to increase the value of the company. This is the incentive that is offered by options. The share price must go up for an option to be worth something. In this way, we align the interests of our employees with those of our shareholders.We have created a trust, that holds a pool of shares available for allotment to our employees they exercise their options.We are growing our Sales and Marketing team, our Engineering team, and business operations to meet the needs of our clients. We must have enough people to support a team of Certified Solution Providers and Certified Design Partners. In parallel to our efforts to market the AKIDA technology through sales channels we continue our social media campaigns, publishing in scientific journals and professional publications, trade shows and direct sales.With the recent elevation of Brainchip into the ASX300 and the OTC QX upgrade in the United States we expect to attract more institutional investors.I will be doing a roadshow in the eastern states after this AGM.We are also improving our investor relations communications and you will notice this going forward. We recently employed a professional Investor Relations Manager, who is based in Perth and speaks Australian rather than techie. His name is now on all our releases and he is happy to answer your questions.We are evolving, not only as an industry leader in Artificial Intelligence hardware, but also as a professional company.We don’t make the sensors, we make them intelligent. Making AI simple – that is also part of our mission. Thank you.- END - Page | 12