Fast Analog Neuron Chip Simulates Human Thinking at Low Power

oshiba Information Systems (Japan) Corporation has developed a new analog neuron chip and has demonstrated its basic performance. The chip realizes information processing close to human intellectual activities. The company targets its future commercialization as it is suitable for internet of things (IoT) devices that require low power consumption operation.

High-Speed Response with Minimal Power
The artificial intelligence (AI) processing represented by a neural network requires a huge amount of operations. It is difficult to operate at high speed with ultralow power consumption in a general-purpose digital computing unit installed with a microcontroller. Furthermore, semiconductor analog circuits tend to become unstable in an environment with ultralow power consumption.

Details of a Developed analog neuron chip
Photo: Developed analog neuron chip
The IoT device is connected to various sensors and needs to perform information processing in real time. It has been expected to realize, using ICs, ultralow power consumption neuron circuits that can respond to input from a sensor at high speed like spinal reflex in living body.

At present, Toshiba Corporation is conducting research and development of analog neuron chips as brain mimic hardware. Based on this operation principle, Toshiba Information Systems developed an IC in collaboration with Toshiba and demonstrated the basic arithmetic operation and processing capability.

The response speed of this IC as a neural network is at more than 10 times as fast as microcontrollers that operate with similar power consumption. High-speed intelligent processing is possible like spinal reflex in living body. Therefore, it can increase the speed of detecting signs of abnormality and failure based on the sensor information of IoT devices with ultralow power consumption. The basic principle confirmed up to now will be further developed for the purpose of realizing its commercialization.