Will this be the world's lowest power AI chip?

Our brains may not be large, but they pack a ton of computing power. For this reason, many researchers have been interested in creating artificial networks that mimic the brain's neural signal processing. These artificial networks, called spiking neural networks (SNNs), could be used to create intelligent robots, or to better understand the brain itself.

However, the brain has 100 billion tiny neurons, each of which connects to 10,000 other neurons through synapses and represents information through coordinated patterns of electrical impulses. Simulating these neurons using hardware on compact devices—while ensuring that the computations are done in an energy-efficient manner—has proven challenging.

In a recent study, researchers in India achieved ultra-low energy artificial neurons, making the arrangement of SNNs more compact. The results were published in IEEE Transactions on Circuits and Systems I: Regular Papers on May 25.

Just as neurons in the brain spike at a given energy threshold, SNNs rely on a network of artificial neurons where a current source charges a leaky capacitor until a threshold level is reached and the artificial neuron fires, and the stored charge is reset to zero. However, many existing SNNs require large transistor currents to charge their capacitors, which results in high power consumption or the artificial neurons firing too quickly.

In their study, Udayan Ganguly, a professor at the Indian Institute of Technology, Bombay, and his colleagues created an SNN that relies on a new compact current source to charge the capacitor, called band-to-band tunneling (BTBT) current.

With BTBT, quantum tunneling current charges the capacitor at an ultra-low current, which means less energy is needed. In this case, quantum tunneling means that current can flow through the forbidden gap in the silicon of the artificial neuron through quantum wave-like behavior. The BTBT approach also eliminates the need for large capacitors to store large amounts of current, paving the way for smaller capacitors on the chip, saving space. When the researchers tested their BTBT neuron approach using 45-nanometer commercial silicon-on-insulator transistor technology, they saw significant energy and space savings.

“Compared to state-of-the-art [artificial] neurons implemented in hardware spiking neural networks, we achieved 5,000 times lower energy per spike over similar areas, and 10 times lower energy per spike over similar areas,” Ganguly explained.

The researchers then applied their SNN to a speech recognition model inspired by the brain’s auditory cortex. Using 20 artificial neurons for initial input encoding and 36 additional artificial neurons, the model could effectively recognize spoken words, demonstrating the feasibility of the approach in the real world.

Notably, this type of technology could work well for a range of applications, including voice activity detection, speech classification, motion pattern recognition, navigation, biomedical signal, classification, and more. Ganguly noted that while these applications can be accomplished using current servers and supercomputers, SNNs could enable these applications to be used with edge devices, such as mobile phones and IoT sensors — especially where energy constraints are tight.

He said that while his team has demonstrated that their BTBT approach is useful for specific applications such as keyword detection, they are interested in demonstrating a general-purpose reusable neurosynaptic core for a variety of applications and customers, and have created a startup called Numelo Tech to drive commercialization. Their goal, he said, “is an extremely low-power neurosynaptic core and to develop a real-time on-chip learning mechanism, which is the key to autonomous bio-inspired neural networks. That’s the holy grail.”

*Disclaimer: This article is originally written by the author. The content of the article is the author's personal opinion. Semiconductor Industry Observer reprints it only to convey a different point of view. It does not mean that Semiconductor Industry Observer agrees or supports this point of view. If you have any objections, please contact Semiconductor Industry Observer.

Today is the 3072nd content shared by "Semiconductor Industry Observer" for you, welcome to follow.