Funding the quest to take AI to mythical proportions

Shepherding powerful machine learning into every-day gadgets with our investment in Mythic.

By Shahin Farshchi, Ph.D.

Artificial intelligence is dominating headlines. Google uses it to predict what we’re looking for and tee up the best results. Facebook uses AI to automatically tag our friends, suggest new ones, and place ads in our news feeds. We funded Nervana to supercharge AI in the cloud, and Intel acquired it to capture a greater share of chips going into servers. We rely on broadband to bring the power of AI from the cloud into smaller, “dumb” embedded devices, such as cameras, wearables, and embedded sensors.

How we use computers typically follows how they work. Power-hungry, unreliable vacuum tubes were conducive to the mainframe computers that mailed out social security checks and ran computational fluid dynamics. Silicon-based transistors made memory and processing cheap enough to go on a desktop, enabling the spreadsheets and word processors that put computers on every desk. Silicon radios and cheaper compute put almost infinite processing power in the cloud and connected it to powerful devices in our pockets. Analog compute will condense data centers into chips bringing advanced intelligence into our gadgets, and our future robotic friends (or overlords).

Unfortunately, sending raw data to the cloud and waiting for an answer isn’t practical. Connections are unreliable, which further dampens AI’s potential for most real-world embedded applications. Uber and Zoox plan to put AI in cars to offer rides without human pilots. They cannot wait for the cloud to tell them to steer or brake, and are challenged with the power and cooling requirements of trunk-mounted servers. Anything smaller than a car is relegated to having to wait for an answer, which is why experts anticipate every AI-powered device to eventually become its own mini data center.

Scientists, engineers and founders have not been sitting idle. Almost three decades ago, Caltech Professor Carver Mead envisioned a future of computers resembling the brain. However, the nature of compute at that time was more conducive to traditional processors. Fast-forward to to today’s neural nets, and it’s becoming more clear that chips mimicking our brains will rule the robotic age. While some startups and large companies are squeezing more efficiency out of conventional digital circuits, we had yet to see some of the interesting analog compute work being done in academia mature into a startup.

That is, until one of the world’s largest car companies came to me lamenting how persistent, high-quality local voice recognition — equivalent to cloud-powered Siri — outstrips available compute resources in their vehicles. On their quest for a solution, they came across an incredibly talented pair of scientists putting the compute power of servers into tiny silicon chips; chips that could go into a steering wheel or turn signal stalk. Those founders were Mike Henry and Dave Fick.

Mike and Dave had previously teamed up at the University of Michigan to find creative ways to cut the power consumed by deep nets, not by 10x, but by something closer to 100x. In doing so, they figured out how to get analog compute to work with today’s AI. Our military asked them to slash the power consumed by part of a GPS receiver for our soldiers, which they did by over 65x. They chose to found a startup that not only frees up soldiers from massive batteries, but decouples the notion of powerful compute from power-hungry computers.

The opportunity, and challenges resembled Nervana: ushering new AI capabilities using novel semiconductor technology. We partnered Mike and Dave up with Nervana founder Naveen Rao. We teamed up with an amazing syndicate of frontier-tech investors, while Mike and Dave are assembling a magical team demonstrating powerful, lightweight implementations of AI that are otherwise impossible with standard digital silicon technology.

Lux is proud to be partnered with Mythic on this epic journey to proliferate and unleash the true power of AI!

Shahin Farshchi is a partner at Lux Capital, where he thinks in analog and invests in hardware, robotics, AI, and space startups. Follow him on Twitter:@farshchi