The most popular and comprehensive Open Source ECM platform
Let us know more
Artificial Intelligence: Programmable Resistors Speed AI Processing
AI researchers increasingly are trying to improve the costs and shrink the time needed to run deep learning computations. One surprising option that is getting more interest is the development of analog processing hardware.
Programmable resistors are the analog equivalent of transistors in digital processors. Resistors can be laid out into layers in patterns that form connections with each other and then run computations similar to how digital neural networks operate.
For example, researchers at the AI hardware company Mythic said that “we use analog computing for our core neural network matrix operations, where we are multiplying an input vector by a weight matrix. Analog computing provides several key advantages. First, it is amazingly efficient; it eliminates memory movement for the neural network weights since they are used in place as resistors. Second, it is high performance; there are hundreds of thousands of multiply-accumulate operations occurring in parallel when we perform one of these vector operations.”
A team of researchers led by MIT identified an inorganic material for fabricating programmable resistors that enable computations to run nearly a million times faster than previous devices. The material can be controlled at the nanometer scale and could be used for devices that can perform deep learning. The result is that deep learning networks will be trained more quickly and cost-effectively.
Murat Onen, postdoc student at MIT, said that “once you have an analog processor, you will no longer be training networks everyone else is working on. You will be training networks with unprecedented complexities that no one else can afford to, and therefore vastly outperform them all. In other words, this is not a faster car, this is a spacecraft.”
