Ever heard of GPUs that can be programmed on the fly? Vividsparks, a company from Hubli, Karnataka, is working on making this possible. Dr. Vijay Holimath, President and CEO of VividSparks, talks about how their company brought a revolution in the computing world.
Q. VividSparks specializes in using the POSIT number system in computing. What market or technical problem do you seek to solve with your product?
A. Let us put it in a simplified form. We all have a smartphone, and whenever you watch a movie, YouTube, or play a game, your phone’s battery goes down drastically. Nowadays, you get fast mobile phones with high-end and high performance, but that speed comes with a cost. Hence, when you require speed, you consume more battery power.
Hence, these problems were addressed in many different ways. Consequently, manufacturing semiconductor chips with a small size in the range of five nanometers, seven nanometers, or 10 nanometers requires lower fabrication and hence lowers the technology nodes, which provides higher speed. As it consumes a smaller chip, your phone or computer becomes smaller. People are addressing the problem in the above-mentioned way. But we tried to address it in a very fundamental way. The fundamental lie in how we compute math in computer people are using the floating point (FP) number system which was invented in the 1980s. This old FP architecture is modified as per the requirements and technology nodes, which results in the design of a new processor. At the same time, we came up with the idea of addressing this battery drain issue by changing the number system.
Computation has a broader perspective, including writing a program on a computer, playing games, etc. There is mathematics behind it. Hence, when you watch a movie or when playing a game, the graphics or lighting effects you see on the phone are due to mathematics such as trigonometric functions, transformational functions, divisional functions, square roots, partial differential equations, probability theory, etc. People are solving this problem with a floating point number system, invented in the 1980s. We came across a novel number system known as the POSIT number system, which was invented in 2017. It took almost 10 years of research and development efforts from 2012 to come across this novel architecture.
We are using a POSIT number system to make the chip smaller and lower the power. Hence, whatever you implement on the chip will be cheaper. In other words, the carbon footprint is lower; hence, power consumption is lower. So, if you run a game, play a game, or watch a movie on your mobile phone or computer, your battery won’t drain drastically. You might now need to charge your battery every five or six hours, even though you may use high-end products while using our system, you may not need to charge your battery for two or three days using our system. So, in the meantime, you’re also offering high performance. Hence, the solution to all the problems is packed into one number system called the POSIT number system.
Q. What kind of end devices do your chips produce?
A. We have numerous end devices. It is not limited to smartphones or mobile phones but also includes high-performance computers and workstations as well. We closely work with various important customers across the globe. These computation units are not limited to laptops or mobile phones it is also used in (IoT) the Internet of Things, Edge computing, Industrial Automation, and also in Self-drive assistance cars. All these things are driven by our processors.
Q. How did you come up with this idea? What’s the story behind it?
A. Initially, it was not about funding a company or making money. It all started when I (Vijay) was sitting at home, very unusual. I was pursuing engineering, and after finishing the practical, I came in the afternoon. I started looking at computer architecture books and began turning the pages. I noticed a floating point unit, which got me interested. For example, whenever we do all this mathematics using pen and paper, it is further fed to the computer in a very simple way to compare and give the results. In the Maclaurin series, Taylor series, or partial differential or Gaussian theories, which you get to learn in engineering degree books or Bcom or BSC degree books, you see all these mathematical equations.
“We are always motivated towards mathematics, mathematics runs in our blood.”
I was always fascinated by how these things we solve using paper just by adding integration zero to infinity and solving them, but how the same thing is solved by a computer. This got me into madness. I started reading a lot of books, after which I came to know the technical term “computer arithmetic.” In other words, you’re doing all this computation on a computer. It is also called computer arithmetic. Hence, after finishing my engineering, I did my MPhil and Ph.D. in the same domain, i.e., Computer Arithmetic. These are R&D degrees, and I published some papers and found novel ideas and novel implementations. After that, I started working on it, and I worked on the same domain in Tokyo. I came back to India in 2014. Then I suppose the POSIT number system changed the whole story.
Q. How did you make your prototype, the first MVP, the proof of concept, first one you made? How did you do it?
A. We closely worked with Prof. John Gustafson. We are the first to prove the POSIT number-based system, and that’s our working prototype.
“We introduced the world’s first POSIT-Based general purpose graphic processing unit (GPGPU).”
Many companies have GPGPUs based on different technologies, but they are not POSIT-based. Hence, we have got the world’s first posit-based processor, GPGPU, that is made in India, and this crown also goes to me. We also have the compiler on this POSIT-based number system as well. It’s also the world’s first compiler. So we have all the prototypes.
Q. What is your company’s main USP for all its products or technologies? One Unique Thing?
A. Nowadays in the semiconductor market, everything is licensed-based. In other words, if you want to develop a processor and want to sell it as a processor, usually people make money when they license their IP cores and sell them to the companies, and then they generate revenue while we make revenue on a subscription-based. For example, in the future, you get a mobile phone for free while paying only $1 per year or only $2 per year, after which the mobile phone is yours. In that case, we generate revenue on a subscription-based model, which is our USP. The products we have consumes half the memory size compared to all the products in the market. Another USP is that our products are programmable, one can program both hardware and software. In other words, if there is a bottleneck in any of your applications, you can change the program software or you can change the hardware, both the contouring, which is unlike in the industry. So these are all our market penetration or go-to-market strategies, or you can also say USP.
Q. As you mentioned, one can change the hardware and software like FPGAs and rewrite the whole software and change the internal circuitry. Can you elaborate on that?
A. The most interesting USP is, e.g., when you buy a blank phone and it does not consist of anything, it only has a memory and scratch card. Then the salesman will ask you about your requirements, such as high-speed, low-speed, or medium-speed. When you tell the sales guy your requirements, he will program your phone accordingly and give it to you. Then you have to pay, let’s say, $1 or $2 per year. Then, after one year, you feel your phone is very slow. It’s kind of a lot of memory. Then you go to the salesman and tell him to make the changes as per your new requirements. He will then change the whole thing. He’ll give you one big file to install, and then it will be like a new phone again.
“Nowadays processors in mobile phones are not programmable and are static, but using our technology you can change both hardware and software.”
This will give you high performance, and then you can run it for a few more years. So this is how the model works. You can have a new phone every year without buying one. On average, a new phone costs around $135 to $200 in the range. If you use our technology, you don’t need to buy a phone every time. Only updating it will provide you with all the new features. Hence, there are a lot of advantages for the companies and the customers, and the carbon footprint is lower, hence its useful nature as well. If you use one phone for 10 years or 20 years, you don’t need to buy a new phone. So the idea behind this is to increase the lifecycle of the product.
Q. Can you tell us about your AI Multicore processor?
A. As AI/ML comes into a lot of things, I’ll give one example. In defense, there is image recognition that comes in. Let’s say the Navy has to detect objects underneath the sea or detect missiles. They have to detect it and have to take exact action. Hence, efficient image video processing is needed, for which artificial intelligence comes in. There are algorithms to make this detection, and a lot of mathematical computation is involved. Hence, the two most important parts are inference and training. So you have to train it and then infer it. In this case, the POSIT has a significant role because it has a higher dynamic range and a lower data width. In other words, with lower data width, you can get good results. So, whatever the technology you see, nowadays, state-of-the-art, the dynamic range is low. In other words, whatever the input or output you do to the circuit, it will either go underflow or overflow. The result has two extremes, that is, if you look at mathematics, there are higher and lower data ranges, i.e., zero to one, one to two, etc. If it goes on the right side, it is greater; if it goes on the left side, it is smaller. Hence, the smaller the numbers go, it underflows; the bigger the numbers go, it overflows. In the current processor, it overflows, and they will chop it off here and chop it up there. Using our technology, there are no overflows or underflows; it remains the same. Hence, higher precision means highly accurate detection and correction algorithms and good mathematical results.
Q.What kind of research and development operations happen in India?
A. We did all the research and development for chips in India, such as RacEr, GPGPU, and FalKon. Thus, all the R & D work was carried out in India, but our engineers are all over the world, like in Madrid, Tokyo, Paris, and Washington. So, our HQ is in India, but our engineers and our main team are across the globe.
Q. What has been your commercialization story?
A. Nowadays people don’t want to pay, which is the hard truth. People don’t want to pay for IP cores, especially in semiconductors. Hence, we decided to have subscription-based sales, so they can use it at a very minimal cost. Thus, using our product, they don’t need any infrastructure, lab, or hardware to run the board, etc. They can use it on the cloud. They can use it, let’s say, for $100 for a year or $100 per month. So, this is how we are doing the commercialization. For our partners, we provide the IP cores free of charge. Thus, they integrate it, and then we charge the end customers the same subscription-based as I mentioned, either monthly or yearly. Our chips are available commercially all over the world.
Q. What kinds of companies are your customers? What kind of people generally approach you for subscriptions in India?
A. Companies, institutes, and R&D centers in AI/ML, HPC, GPU computing, and automotive computing. In India, we closely work with government institutes. Usually, abroad, we are approached to accelerate their applications. They want to see how much more accurate, how much better performance, or low memory consumption they can get by using our product. Thus, people are very excited about using this technology.
Q. Are you currently looking for funding or investment for your company?
A. For now, we do not require funding, but if we have a good opportunity that is mutually beneficial, we can consider it. But I’m not so keen on the investment, as we have good customers providing us with a good profit.
Q. What are your general hiring trends?
A. Since our field is highly technically specialized, some of the students are not so equipped with these kinds of technologies, i.e., semiconductors. Students are well-versed with VLSIs, VERILOG, and FPGAs, but in our case, it requires high mathematical skills. Research professionals, maybe we plan to provide internships.
Q. What are the plans for the expansion of ViviSparks in India and abroad?
A. In India, I don’t require any branches because only HQ is enough, but we are planning to expand to other countries. We are mainly an export-oriented company, so we don’t focus on India alone. We have strong distribution across the globe.
Q. Are there any opportunities for other firms to partner with you to expand your customer base?
A. Ya, of course, we are excited about partnering with any institutes or companies. If it is for providing funding and is beneficial for them or mutually beneficial, then we would love to be partnered.
Q. What do you want to add or convey to our audience?
A. I would add that our technology is subscription-based which will provide phones or any digital equipment at a cheaper cost. Hence, we believe technology has to be novel—not expensive. And novel technology should be accessible and affordable for everyone.