AI could get major speed boost from new type of DRAM, research groups sayNews
February 08, 2021
IEEE INTERNATIONAL ELECTRON DEVICE MEETING 2020. Separate research groups in the U.S. and Belgium believe that a new kind of dynamic random-access memory (DRAM) could hold bits hundreds or thousands of times longer than commercial DRAM and enable huge savings in space and energy when used for artificial intelligence (AI) and large neural nets.
Presenting at December's IEEE International Electron Device Meeting (IEDM), Arijit Raychowdhury, a professor of electrical and computer engineering at Georgia Tech who worked with researchers at Notre Dame University and the Rochester Institute of Technology on a new embedded DRAM, said that “The challenge with the monolithic 1T1C design [single-transistor, single-capacitor] has always been the difficulty of building the capacitor as well as fabricating a transistor with ultra-low leakage,” using a manufacturing process meant for logic transistors.
Read and write operations turn the transistor on and off; while this system is very fast, very cheap, and power-efficient, it does have some disadvantages: For one, reading the bit drains the capacitor, so reading means writing the bit back to memory. Additionally, even if you don’t read the bit, charge will eventually leak out of the capacitor through the transistor, which means that all of the cells must be periodically refreshed in order to keep the data. Today's DRAM chips refresh every 64 milliseconds.
In contrast, the new embedded DRAM is instead made from two transistors only, no capacitor (2T0C), a process that works because a transistor’s gate is a natural -- albeit small -- capacitor, so that the charge representing the bit can be stored there. Raychowdhury says that the 2T0C arrangement does not work well with silicon-logic transistors, as any bit would drain away immediately because the transistor gate capacitance is too low and the leakage through the transistors is too high, which led the researchers to use devices made with amorphous oxide semiconductors, like those used to control pixels in some displays.
Researchers at Belgian research and development organization Imec debuted a similar 2T0C embedded discovery at IEDM, with the Belgian team using indium gallium zinc oxide as a semiconductor.
Additional details can be found in the IEEE Spectrum.