Scientists Create a 3D Chip Capable of Making A.I. Systems Work

Imagine a world where A.I. is all around you. You get in your self-driving car to go to the doctor’s office, where a slew of tests are analyzed by machines that can diagnose your ailments with 99 percent accuracy. They give you a personalized prescription based on your individual biology, and then you go have a lunch of a cheeseburger and a salad, one catering to your tastes and the other to your needs. Maybe you cheat and get fries with the cheeseburger, anyway.

For machines to accomplish that kind of work, they need the type of hardware that can handle the massive amount of data required. That’s where researchers from the Massachusetts Institute of Technology and Stanford University come in, as they recently developed a new type of three-dimensional chip made from different nanotechnologies that essentially puts the main two functions of chips under one roof. The chip streamlines the process and makes it easier for systems built from this chip to function as prescribed for A.I. systems.

Conventional chips basically come in two different flavors — those for data storage and those for processing, and they need to be linked in order to make the system run. In a paper published this month in the journal Nature, the research team outlines a new design for a chip that cobbles together both these functions.

The new chip is made of carbon nanotubes (sheets of 2D graphene morphed into nanocylinders) and resistive random-access memory (RRAM) cells, which charge the resistance of solid dielectric materials.

It might sound a bit complex, but what it basically means is that the RRAM and carbon nanotubes are stacked vertically over one another, creating a 3D architecture that lets a single chip fulfill multiple functions. This is beyond the capabilities of silicon-based chips.

Computers made with such a design could handle incredible amounts of bandwidth — the type we’re likely going to need in complex computing structures that use A.I. and autonomous systems. Any machine-learning applications would likely get a boost from a such a chip.

“The technology could not only improve traditional computing, but it also opens up a whole new range of applications that we can target,” said lead author Max Shulaker in a statement. “My students are now investigating how we can produce chips that do more than just computing.”

The team is far away from demonstrating how the chip could be viably used in real world devices. But the fact that A.I. is still a work in progress gives the team plenty of time to figure out a sustainable way to manufacture and implement this chip in industrial and commercial applications.