Edge-processing. It’s as important as it is cool-sounding. With edge-processing, data processing happens close to the source, eliminating the need to send data to a remote location — such as the cloud — for processing. This means data is handled at the “edge” of a network, inside a device, or at a local server. This allows devices to compute, analyze, and store information locally, removing the cost and lag time of moving raw data to processing locations that could be thousands of miles away. This high-speed edge-processing can be handled by a new smart AI chip that consumes little power and is inexpensive to make.

The Japanese AI startup ArchiTek’s AiOnIc processor is designed for drones, robots, autonomous vehicles, VR headsets, and other devices. Its makers say “AiOnIc processes data more accurately and faster than any other leading-edge AI chips can, at a fraction of their cost and power requirements.” ArchiTek says its chip consumes under 2 watts of power and has a production cost of about $10 per chip. Plus, it’s a “one-chip solution,” able to handle the AI processing for peripherals, including a PCU, Ethernet controller, SD card controller, and more.

What about performance? For that, we should first talk about TOPS, the acronym for Trillions of Operations Per Second. TOPS tells you how many computing operations (or basic math problems) an AI chip can handle in a second. ArchiTek claims its chip performs more than 6 TOPS (over 6 trillion computations per second) and is powerful enough for real-time operations in a variety of devices. The chip can handle simultaneous computing with several AI algorithms at once. All in a small package: the chip measures just 12 millimeters by 12 millimeters.

ArchiTek CEO Shuichi Takada says, “AiOnIc consistently does a superb job at incredible speeds on battery power, which is unheard of for high-performance edge-AI chips…we believe it can make autonomous vehicles and intelligent machines more efficient and revolutionize productivity when we use them in everyday life.” For health tech applications, this could mean inexpensive, smaller, lower-power wearables that can process sensor data immediately and only pass along the useful information that they contain.