Applied, ARM to develop CeRAM for neuromorphic applications

July 31, 2018 // By Peter Clarke
Semiconductor manufacturing equipment maker Applied Materials Inc. has been selected by DARPA to work with ARM Ltd. and research firm Symetrix Corp. to develop a switch that functions like the neuron and synapses of the human brain, based on Correlated-Electron RAM (CeRAM) technology.

The project is part of DARPA's Electronics Resurgence Initiative (ERI) to develop new computing materials, designs and architectures. ERI has a budget of hundreds of millions of dollars to be spent over the next four or five years. The goal of the particular project is to speed up neural network processing while improving power efficiency through the use of analog signal processing as compared to current digital approaches.

The CeRAM – or Correlated-Electron RAM – is a form of non-filamentary, non-volatile memory based on metal-insulator Mott transition in nickel oxide and other transition metal oxides (TMOs). Because the switching effect is based on state transition throughout the material rather than the migration of ions or vacancies it is expected to have superior endurance and reliability. Carlos Paz de Araujo, a professor at the University of Colorado who is the leading advocate for development of the memory, has led much of the work on the technology through his company Symetrix Corp. (Colorado Springs, Colo.).

In 2014, when Professor had already been working on the technology for about five years, the claim was that with nickel oxide the state was robust up to 400 degrees C and can be read with a voltage of 0.1 to 0.2 volts. Devices albeit at large geometry were reported to have 10^12 cycles read endurance.

In 2014 Professor Araujo secured support from ARM for continued research into CeRAM as an embedded non-volatile memory that could potential have superior properties to flash memory and the various two-terminal ReRAMs that are in development or coming to market (see ARM's turn to non-volatile memory is right move). If CeRAM is truly bulk switching it could have a scaling advantage over filamentary ReRAMs that may struggle to go below 10nm just as flash memory struggles to scale below 28nm.

Next: DARPA recognition


Vous êtes certain ?

Si vous désactivez les cookies, vous ne pouvez plus naviguer sur le site.

Vous allez être rediriger vers Google.