Processing in memory: a promising solution to Moore's Law limits?

Scientists claim that processing on ReRAM memory will allow smaller, faster, more energy efficient devices.

A group of scientists from Singapore and Germany have announced a promising development in the quest for producing smaller and more energy efficient computing chipsets.

Moore's Law is approaching a hard limit when it comes to traditional microcircuitry. It is becoming harder, physically, to produce silicon wafers that are smaller than the previous generation, as eventually the atoms that make up the transistors become too close together to work.

This has led scientists to look at new ways to speed up processing. One is quantum computing which will be revolutionary when it arrives, but which is some way off as a general purpose computing platform for everyday use.

Another approach is to look at memory. A new technology called Redox-based resistive switching random memory (ReRAM) should be commercially available soon. ReRAM is a non-volatile RAM that works by changing the resistance across a dielectric solid-state material. It has a good long-term storage capacity and can be produced at nanoscale. It promises to increase I/O speeds while also reducing power consumption. ReRAM is being developed by a number of major semiconductor vendors.

Assistant professor Anupam Chattopadhyay from Nyang Technology University, professor Rainer Waser from RWTH Aachen University and Dr Vikas Rana from Forschungszentrum Juelich have discovered that ReRAM is not only useful as an alternative to traditional DRAM; it can also be used as a processing platform, meaning data does not have to be moved to and from a CPU. Processing data in the same place it is stored is far more efficient, allowing for faster and thinner mobile devices.

Another feature of the prototype circuitry should also allow for quicker processing. Rather than operating on the familiar binary system (0,1), the ReRAM based circuitry being developed by the scientists stores and processes data using a quaternary number system (0,1,2,3). This should increase the processing efficiency because a quaternary number is shorter than its binary equivalent. Chattopadhyay explained that in current computer systems, all information has to be translated into a string of zeros and ones before it can be processed.

"This is like having a long conversation with someone through a tiny translator, which is a time-consuming and effort-intensive process," he said.

"We are now able to increase the capacity of the translator, so it can process data more efficiently."

Professor Waser explained that the new system is promising for the development of future IoT and wearable devices.

"These devices are energy-efficient, fast, and they can be scaled to very small dimensions," he said.

"Using them not only for data storage but also for computation could open a completely new route towards an effective use of energy in the information technology."