SiMa.ai is a machine learning company enabling high-performance compute at the lowest power. Initially focused on solutions for computer vision applications at the embedded edge, the company is led by a team of technology experts committed to delivering the industry’s highest frames per second per watt solution to its customers.
Here is the exclusive interview with the leadership team of SiMa.ai conducted by Vaishali Umredkar, Electronics Maker.
- KRISHNA RANGASAYEE | Founder and CEO, SiMa.ai (Leading the interaction)
- GOPAL HEGDE | SVP, Engineering and Operations, SiMa.ai
- SUDERSHAN R. VURUPUTOOR | Sr. Director, Hardware Development, SiMa.ai India Pvt Ltd in Bengaluru
- AMIT KUMAR MITRA | Sr. Director, Software Development, SiMa.ai, Bengaluru
What do you mean by is your ML green?
Krishna: We have seen a lot of attention paid to performance and machine learning in the last few years and as we well know, the carbon footprint that’s consumed for computing on the cloud and the edge is skyrocketing. It is projected that in the next five years, data centers are going to consume about 6% to 7% of the carbon footprint of the entire planet. This is enormous. If you really look at it, 10 – 15 years ago this wasn’t even available as an entity. It’s really becoming as significant as automobiles in terms of carbon footprint. So again, a lot of attention is paid to performance and people are not worrying about power and the carbon footprint that’s being generated in the data centers. We are a company focused on enabling high-performance compute at the lowest power for the embedded edge and this is where I think machine learning is really going. This will affect lives in a bigger, more profound way in the next decade or two. And obviously, given that the edge is much bigger than the cloud, if we don’t get ahead with power-efficient machine learning solutions now, it is going to be an even bigger problem later on. We are very passionate about addressing this and we have built the company around building a very efficient architecture for both power and performance efficiency. We are currently leading the industry by at least 20X in that aspect. Technologically, we know it can be done and we believe that it’s a responsibility to manage the power efficiency. We didn’t pick this tagline for us — our customers actually did because we are one of the rare companies that’s addressing power consumption and are pushing for machine learning to be green and sustainably conscious.
Tell us about the new Bengaluru Design Center and what it brings, especially in India?
Krishna: Bangalore is a well-known entity and very familiar to us. Beyond the familiarity, I believe that it is a very rich talent pool of people from great universities, with great multi-national experiences and with the right hardware, software and now machine learning technical expertise in addition to an increasing knowledge and expertise of systems. We have a really good team, mostly based in the Bay Area in California. As we looked at our global presence, it became very obvious that Bengaluru would be a good start for us to really tap into the great talent pool in the region. We also know that it’s not just a development opportunity alone; India is a market itself using machine learning in many different ways. We want to have a local footprint so we can understand and learn and also see how we could play our part in contributing to India’s growth from a machine learning industry adoption standpoint.
Gopal: I have a big affinity for Bangalore. The tremendous talent in Bangalore has grown over the years in terms of the sheer volume of highly talented silicon designers, software engineers, embedded software engineers, machine learning compiler experts, and people proficient in toolchain, test and release engineering. All of these skills are found in abundance and Bangalore also has a very strong startup culture. Many of the world’s leading startups were formed in Bangalore. We see opportunities to ramp up our team very quickly there and bring on board quality engineers with exceptional expertise.
Amit: The kind of problem we are solving at SiMa.ai is something which is really complex. Such a complex problem requires a certain level of expertise in hardware and physical designers to experts in deep learning. We have that kind of ecosystem and expertise in Bangalore. We have people who have been working right from the physical layout of hardware designs to deep learning architectures. That’s a kind of a unique combination we were looking for and what makes Bengaluru the best choice.
Are there any specific applications targeted for India versus other markets?
Krishna: As a company globally, we’re focused on three key application areas.
One key area is what we call smart vision. This includes security, safety, surveillance, medical scanners, and broad imaging applications that are used for both safety and security. The second one is robotics which is also a very broad category. Our initial focus is really on factory floor robotics but we are scaling our robotics focus globally into lots of different areas. The third area is automotive and autonomous systems. So smart vision, robotics, and autonomous systems is our global focus and is the same in North America and Europe. These are the key areas we are initially bringing into India and I think it’s fair to say we are beginning to learn more about the needs of the Indian market.
India itself may represent new opportunities and new applications. We think of ourselves as a computer vision platform that could solve more problems and address more needs than just the three applications we’re initially focused on. Obviously, as a startup, we need to be disciplined to make sure that we are focused on these three but we are going to take the time now and really learn a lot about the needs of the Indian market. We realize there’s a big opportunity for us to learn from India and do something different that can scale globally as well.
Machine Learning is of great importance today. How are you bringing this technology to industry applications?
Krishna: The focus for us as a company is really on the embedded edge market. Today, it’s serviced by classic SoC companies — or system on chip companies with about $40 billion dollars consumed on an annual basis. If you really take a step back though, the existing technologies servicing the embedded edge market (smart vision, robotics, and autonomous systems) are at least 30 years old. These products have been around for a long time, they’ve been designing for a long time, and they’ve been in production for a long time. I think what’s happened is the amount of data and the amount of video content has far exceeded the compute capacity of these classic devices. That’s the problem that the industry faces and machine learning can help. The industry startup response has really been to offer machine learning accelerators and there’s been a lot of companies created. Customers, however, do not want to walk away from legacy applications to widely adopt machine learning though. So, our number one priority is to build a purpose-built platform and we call it our machine learning SoC or MLSoC™. It does three things very well. First, it delivers very high-performance power, which is at minimum 10 – 20x better than anybody else so we can meet the demand of today’s new data and video needs. Second, it delivers a software experience which is very, very easy. A push-button experience in software is extremely important for scaling machine learning. Lastly, since we are purpose built, we can support classic SoC legacy, along with machine learning in one unified offering. So really, the highest priority for us as a company is to scale machine learning from nascent market and PLC market elements into full production systems, and really help the industry adopt this technology and displace the old technology. That’s an enormous market opportunity and we feel like we are one of the unique companies addressing the core big issues.
Gopal: If you think about machine learning today in the embedded edge market, whether it’s robots, smart vision, drones, or high-end computer surveillance equipment, these kinds of applications are mostly served by embedded processors with very limited machine learning capability. Many existing machine learning players try to deliver the solution they designed for the cloud now for the edge. Solutions from Nvidia and Intel, for example, are designed for the cloud and they are scaled down for the edge. The ability to support legacy applications that are deployed at the edge is compromised because you can’t use an accelerator to support legacy applications, you need a compute element for that and significantly lower power.
Can you comment on your expected growth rate in coming years?
Krishna: We are a little company taking baby steps and growing but we are tackling a $40 billion market at a very aggressive rate. We have not publicly stated our ambitions in terms of where we are going to be by revenue and market share, but we are already recognized as one of the most innovative companies by the way we are approaching our problem solving. In five years, we will definitely be recognized as one of the significant revenue leaders in this market.
What are your views on India’s engineering talent? And how are you encouraging this talent to grow in your company?
Krishna: India is a very unique and special place, Bangalore in particular, though there are other cities like Hyderabad, Pune and Delhi that also have a lot of great talent. For SiMa.aiand particularly from a machine learning context, SoC context systems understanding Bangalore is really we will be investing quite heavily in scaling up our site there. We are starting small with our new design center in Bengaluru and will have around 20 people before the end of this year, but we definitely think that this is going to be a large site for us. As we grow our success and our team, I have no doubt that this will probably be the largest site. It really comes down to how we scale and the talent pool that we have. We are going to be attracting a lot of key talent not only from universities but also from other multinational companies. We are doing exciting stuff and already we have seen a lot of demand from people wanting to come join us.
Sudarshan: I think one thing that is required for growing talent is to attract a strong leadership team first. As Gopal and Amit mentioned, this particular thing requires multiple skills and talents, not just by ML/AI experts but also other key roles like our computer architecture, computer vision SOC design compiler drivers. As you can see, multiple domains are required and these talents are abundantly available in bandwidth in India. The key is to get the right people and the right leadership team to form a solid foundation first. Then we can actually start growing organically because we have more than 250,000 engineers and qualified engineering graduates every year from India. To actually attract that level of talent and organically grow while also participating in the initial design is very important.
Any comment about India as a semiconductor manufacturing destination?
Krishna: In my mind, India has tremendous potential for us. We are starting with development being a priority right now and we’re going to continue monitoring not only the market opportunity, but also the manufacturing capability. And indeed, as we think about scaling, will be thinking about absolutely everything. We have a strong leadership team in Bengaluru and we’re going to learn from them in addition to our great customers and partners.