What Hiddenite Could Mean for Neural Networks


March 7, 2022

Shop Now

Here at Wafer World, we’re always excited to look into the latest developments in the IC space. It’s incredible how a small change – such as using an InP wafer instead of a silicon one – can have such a large impact on performance.

This is why we’ve watched the development of the Hiddenite chip with such interest. In line with the latest AI boom, Tokyo researchers have developed an IC that reduces the energy and time required of a typical neural network setup.

InP Wafer

How Neural Networks Work

To understand what makes Hiddenite different, it’s first worthwhile to discuss how a neural network typically operates.

Essentially, the AI takes a large variety of parameters and trains with all of them, seeing how they perform over time. This is a great way to come up with effective, sometimes counterintuitive solutions to problems, since AI is going through the steps so many different times.

That said, to work properly, a neural network needs:

  • Power
  • Time
  • Many Examples of a Situation

What Makes Hiddenite Different

Instead of training a variety of different parameters, Hiddenite looks at initial outputs to determine which of the parameters are successful earlier on. It’s essentially not going through the time and energy to train with less effective parameters.

For our purposes, it’s also interesting to look at the architecture of the chip, which minimizes the power necessary by reducing the need for external memory.

Looking for an INP Wafer Manufacturer?

At Wafer World, we offer a variety of wafers, designed to fulfil different computing needs. Whether your company needs INP wafers to lower energy usage or you’re looking for a silicon wafer, you can rest easy knowing that we’re ready to provide what you need.

If you can’t find a wafer on our site, please don’t hesitate to contact us. We’re happy to answer any questions you may have, and we also offer free quotes.

Wafer World Banner