The Other Side Of Artificial Intelligence — Next Generation AI Specific Chips

Most of the attention that AI has received over the last decade has only been on the software side, but the ‘Silicon’ side of things have not received much attention, even though there was (still) a pressing need for better chips and significant developments occurring in this space. Pioneers like Yann Le Cunn and Andrew NG have always focused on both sides of the coin i.e. hardware + software. Better hardware is required to enable shorter training times, better memory access for deep/machine learning algorithms while working with bigger sized training data sets, as well as shorter prediction times all the while minimising error rates. GPUs had previously brought efficiencies ranging in between 9X to 72X over conventional CPUs (Source). Better hardware and chip architectures could bring in a further factor of 10X improvement in efficiency.

Artificial Intelligence Accelerators — The “Silicon” Journey

But behind the scenes, most of the technology giants realize the need and market opportunity for creating dedicated chips (AI Accelerators) for AI applications. Both Google and Microsoft are known to have used custom ASICs to enhance machine learning capabilities of their respective search engines. Intel has made its priorities clear with the $16.7B acquisition of FPGA maker Altera, and the recent ~$400M acquisition of Nervana Systems, a startup working on custom ASICs for deep learning. These acquisitions have given Intel enough ammunition to compete with Nvidia, which is currently dominating the deep/machine learning hardware space.

AI Accelerator Projects of the Tech Giants

The priority for most of these giants is to produce an AI specific chip which could power the next generation high performance data centers. These chips could further power the next generation of IoT devices (Smart Phones, Smart Cars, Smart Toys, Robots etc.) and provide them with in-device AI capabilities, making them less dependent on a remote cloud infrastructure.

Where are the Startups?

Startups have not been far behind in the race. The space has seen an influx of $589M of venture capital and private equity. Knuedge, founded by a former NASA head recently came out of stealth mode after a decade, with a $100M funding announcement. Nervana Systems was acquired recently by Intel for ~$400M. Mobileye, a company which powers the ADAS of prominent car companies like BMW, Volvo, GM, Renault, Tesla(until recently) is leading the way in the self driving car space.

AI Accelerators — Some Prominent Startups

* Read as Startup Name(Founded Year, Founded City, Funding, Company Stage)

** Shows only a few prominent investors.

Academia is also not shying away from this space. Recently, MIT researchers had presented Eyeriss, a chip which was 10X as efficient as a mobile GPU, and could run deep learning algorithms locally on a mobile phone.Going further, we can expect more activity from established companies as well as startups in this space. These developed chips powering next generation IoT devices and data centers.

For more interesting articles visit blog.tracxn.com