With ever-smaller smartphones now flaunting the computing powers of yesteryear’s hefty business machines, it might look like Moore’s Law has yet to be repealed. In 1965, Intel co-founder Gordon Moore saw computer chips doubling in processing power every year, and predicted the exponential rise of smaller, cheaper, and faster devices.
But Michael Henry, co-founder of the artificial intelligence startup Mythic, says we haven’t been sustaining Moore’s constant upward curve recently by steadily shrinking the size of transistors on a computer chip—the trend Moore observed in 1965. Instead, we’ve had an outside assist by offloading a big chunk of the computing tasks of our little devices onto distant servers, often loaded with powerful graphics processing units (GPUs). In other words, we’re delegating our calculations to the cloud.
Henry wants to bring that computing power back to earth, by empowering individual devices including cameras, wearables, and drones with a novel, speedier computer chip that departs widely from standard digital computer architecture. Mythic revealed the basics about the design of its chip today, and announced a $9 million Series A fundraising led by DFJ, which was joined by Lux Capital, Data Collective, and AME Cloud Ventures. DFJ’s Steve Jurvetson and Lux Capital’s Shahin Farshchi are joining Mythic’s board.
Mythic, based in San Francisco and Austin, TX, says the chip’s capacity would allow a single small device to carry out the computations involved in artificial intelligence analysis, instead of forcing it to export the raw data via the Internet to powerful external computer systems for interpretation. It’s early days yet for the small company, but among the future products that might result are drones that can single-handedly scan a railroad track, interpret the images, identify broken bolts, and send a quick bottom-line warning to the railroad operator. With the power of such “local AI,” an autonomous car that lost its Internet connection could still navigate obstacles on a highway, because it would have autonomous intelligence.
But that isn’t possible without a radical revamp of computing processes, Henry (pictured above) says. The pace of chip miniaturization has already slowed, and artificial intelligence software—with its elaborate algorithms that mimic human cognition—has also leapt ahead of industry’s capacity to incorporate that technology into smaller, cheaper devices, he says.
“AI upset Moore’s Law’s scaling,” Henry says. “AI applications have to run on desktop GPUs.”
With Mythic’s new chip design, Henry says, the company can fit the equivalent of desktop GPU computing power into a unit the size of a shirt button. A data center’s powers could fit into a unit the size of a deck of cards, he says.
Beyond that, the chip design could extend battery life 50-fold, and support products that are more secure because data wouldn’t need to be shared with external servers, Mythic says. For example, a voice-enabled home assistant such as Amazon Echo wouldn’t have to send a voice file to an outside transcription service to understand a homeowner’s command. A child could confide in a talking doll without the danger that hackers could capture the conversation by infiltrating a Web-based data storage system.
“It’ll be secure and trusted when it’s local,” Henry says. Mythic plans to test its big claims soon, with field trials involving industry partners looking to solve specific problems. The startup has been adding production-level team members to its crew of R&D engineers, recruiting from companies such as Oracle and Texas Instruments.
Mythic isn’t the first to try to make local AI a reality. It has company in the race, including Boston-based Neurala and Seattle-based Xnor.ai.
Henry and co-founder Dave Fick drew on their work at the Michigan Integrated Circuits Lab at the University of Michigan to form their company in 2012. Its original name was Isocline. The partners wanted to overcome “an unfixable bottleneck” inherent in traditional digital computers, which are based on a dual structure: processors that carry out computing chores, and that communicate with memory that contains programs and stores intermediate results, Henry says. The team eliminated processors by moving to what it calls “hybrid digital/analog computation,” and shifting the computation chores to the memory elements.
Mythic makes chips through standard manufacturing processes, and then modifies them, Henry says. “We pack the chip with flash memory,” he says. Data can be run through existing AI algorithms, with some minimal modifications, Henry says.
Mythic will sell chips, and it will also help customers solve process challenges by maintaining libraries of solutions for various types of tasks, such as detection, voice-to-text transcription, mapping and navigation, and virtual reality. Its first target customers will be involved with products such as smart home devices, action cameras, health-related wearables, security, and industrial drones. In the future, Mythic plans to pursue opportunities in robotics, driverless vehicles, augmented reality, and virtual reality.
On top of its new funding and chip design announcements, the startup also made its debut under the new name chosen collectively by the company team: Mythic.
“We see it as one level above ‘Epic,’” Henry says. “We wanted to show investors something so powerful it doesn’t appear to be real.”
Photo courtesy of Mythic.
