The journal Nature publishes the research results of IBM "AI Analog Chip", which can achieve 14 times the efficiency of traditional chips.

CTOnews.com, August 24 (Xinhua)-- the journal Nature published the latest research from IBM Research Lab on August 23: an AI analog chip that is 14 times more energy efficient than traditional digital computer chips.

It is said that the chip is more efficient than the general-purpose processor in speech recognition, and is expected to break through the bottleneck caused by the current AI development due to lack of computing performance and low efficiency.

The summary of the article says that the current AI model with billions of parameters can achieve high precision in a series of tasks, but it also highlights the inefficiency of traditional general-purpose processors (including graphics processors, central processing units, etc.). To this end, the research team proposed a scheme of "simulated memory computing", which provides greater energy efficiency by performing matrix-vector multiplication in parallel on its own memory.

The research team also developed a 14nm analog chip, which contains 35 million phase change memory units in 34 modules. In the testing process, the research team used Google voice commands and Librispeech speech recognition to test the efficiency of the chip's language processing capabilities. After using Google voice tests, the chip found that the chip's performance and accuracy are "comparable to current digital technology". On the larger Librispeech, the chip can achieve 12.4 trillion operations per second per watt, up to 14 times the performance of traditional general-purpose processors.

CTOnews.com attached the original article of Nature magazine as follows:

Referenc

An analog-AI chip for energy-efficient speech recognition and transcription