In the last three years, starting from the third quarter of 2017 to be more accurate, all of a sudden venture capitalists started funding fabless semiconductor start-ups, mostly in AI and ML.
Bengaluru: In 1986, Walden Rhines, CEO Emeritus, Mentor Graphics, featured on the cover of highTechnology, an emerging technology US magazine in the lead article “AI: Heavyweights get into the Act”, along with George Heilmeier. Both were working in Texas Instruments at that time.
Recalling this incident, Rhines said, “There I was, 32 years ago, talking about artificial intelligence and working on it too. But it didn’t turn out to be a very good business because there was a lack of data which could be analysed – there was no Internet or IoT or social networks to collect sizable data sets. Computing power was also very limited and there was a need for more advanced algorithms. But the single biggest thing, there was no obvious killer application that you could make money with.”
Now, a completely different environment exists and each of those stumbling blocks has disappeared.
“In the last few years, all of a sudden, we have a new environment and the computational power too has dramatically increased. Artificial intelligence and machine learning (ML) need a different architecture, something different from the traditional Von Neumann architecture which no doubt has served us so well. Applications like floating point processing prove that machines clearly outperform humans but in many other areas, humans are still way ahead. In fact, technology-wise, we are orders of magnitude away from being able to match the human brain in a variety of functions including pattern recognition and power dissipation,” he pointed out.
With neural networks becoming fundamental building blocks for AI-related machine learning, a large number of the chips that are coming out today have embedded neural networks to do deep learning and to store non-volatile memory.
“Neural networks are affecting us everywhere. In 2017 more than 300 million smartphones were shipped with some form of neural-networking capabilities. In 2018, around 800,000 AI accelerators were shipped to data centers and every day, 700 million people now use some form of smart personal assistant like an Amazon Echo or Apple’s Siri. So it is becoming a part of what we expect in life,” Rhines pointed out.
VC funds for fabless semiconductor firms
The strange thing is that in the last three years, starting from the third quarter of 2017 to be more accurate, all of a sudden venture capitalists started funding fabless semiconductor start-ups, mostly in AI and ML.
For the previous 20 years, the funding for fabless semiconductor companies had been declining down to a point where it was less than $400 million a year. In third quarter 2017 and in 2018 it jumped, it set an all-time record of $ 3.2 billion of venture-funded fabless semiconductor companies where more than 40% were for AI chips.
Even the dotcom boom didn’t throw up this kind of an investment in fabless semiconductor domain, Rhines noted.
“If we look at VC investment over the last 18 years, the dotcom boom resulted in $2.5 billion per year for VC-funded fabless semiconductor companies. Since then, it has been downhill. It dropped down to around $1.7 billion in the next 5-6 years (2002-2008), steadily kept on decreasing and reached a new low of around $400 million in 2016.
“But then, the strange thing is there was a sudden surge in 2017 where it shot up to $1.4 billion and in 2018 it hit an all-time high of $3 billion. The VC industry which was enamoured of social networking and ecommerce sectors suddenly became interested in the fabless semiconductor industry.”
“Interestingly, if you analyse the type of funding for these companies you notice that it is the first round of funding (a high-risk funding) that has occurred in the majority of the start-ups. The VC was taking a much higher risk because the start-up doesn’t have anything to give or show at that point of time. Which throws up a big opportunity for the semiconductor industry and by the way a big opportunity here in India too,” he added.
India’s strengths in AI
The number of fabless start-ups in India is growing and is driven very heavily by applications that are based on non-traditional computing and that involve some form of alternative processing like AI
“I personally think that India has terrific software capability. Up until 2017, it was mostly software but now what I think is happening is that Indian engineers are doing both hardware and software because traditional architectures are not powerful enough to process all the data you need for AI.”
According to Rhines, there was a lot of interest being shown from Indian companies and start-ups. "I looked at LinkedIn data as to how many people identify AI as their primary area of expertise and I saw that it was the largest number in the US, China came in second with India not too far behind not far behind is India."
And, these three as a group are way ahead of number 4, which is Israel followed by Germany.
“If you look at the US start-ups in AI chips being funded, more than half of them are started either by Chinese or Indians living in the US. There is nothing intrinsic that says India is good at software and not hardware, but the key here is to be good at innovation, integration and development of IC design” Rhines added.
India’s Design Activity Accelerated by Investment Activity
In December last year, the Karnataka government launched the Semiconductor Fabless Accelerator Lab (SFAL) which has charted out an ambitious plan to accelerate 20 start-ups in next three years and 50 in next five years. SFAL also plans to support existing fabless companies with a goal of at least 2-3 products over the next two years.
In Hyderabad, the first Fabless Chip Design Incubator (FabCI) was launched in 2018 by IIT Hyderabad. This incubator for fabless chip design start-ups funded by Ministry of Electronics and Information Technology aims at incubating at least 50 ‘Make-in-India’ chip design companies
Apart from this, the IITs too are supporting AI education.
For instance, IIT Hyderabad offers a master’s degree program in AI and is introducing an undergraduate program in AI during the 2019-2020 academic year, while IIT Kharagpur announced the launch of a six-month AI certificate course as well as a center of excellence for AI in Hyderabad.
The faculty at IIT Madras, home to the Robert Bosch Center for Data Science and AI, has launched a start-up to offer cost effective training in AI for the Indian workforce.
— Sufia Tippu is a freelance tech journalist based in India contributing to EE Times India