Researchers Develop First Working Memprocessors, Plan Future Ultra-Fast Memcomputers
Researchers from San Diego and Turin have developed the first working “memcomputer” prototype, with brain-inspired “memprocessors” able to simultaneously perform computation and memory operations, which could open the way to ultra-fast future computers.
The researchers, headed by Massimiliano Di Ventra, a professor of physics at the University of California, San Diego (UCSD), have published their work in Science Advances. The research paper, titled “Memcomputing NP-complete problems in polynomial time using polynomial resources and collective states,” is freely available online.
Ultra-Fast Brain-Inspired Computing
Memcomputing, inspired by neuronal information processing in the brain, is defined as computing and storing information within the same units, interacting memory cells called memprocessors. The theoretical concept of memcomputing has been around for some time, but the researchers claim that theirs is the first working implementation.
Traditional computers use separate computing and memory units, and therefore must shuffle a lot of data back and forth, which consumes power and time. The brain doesn’t shuffle that much information, which is one of the reasons it’s much more energy-efficient than current computers.
“To make a quick comparison: our own brain expends about 20 watts to perform 10^16 operations per second,” Di Ventra explained to Popular Mechanics, while a supercomputer would require 10 million times more power to do the same number of operations. “A big chunk of that power is wasted in the back and forth transfer of information between the CPU and the memory.”
A memprocessor operates like a transistor, but also changes some of its physical properties, such as its electrical resistance, depending on incoming data and instructions. Therefore, a memprocessor can perform computation and data storage simultaneously.
The prototype memcomputer could prepare the way for Universal Memcomputing Machines, or UMMs [arXiv preprint], which are interesting because they could theoretically perform certain class of computation much faster than conventional computer. So much faster, in fact, that they can be compared to the much hyped quantum computers.
The researchers have used their simple prototype, which has six memprocessors, to solve a complex mathematical problem and demonstrate that, in their memcomputing architecture, increasing the number of memprocessors could permit to tackle problems whose computational complexity increases much faster. The researchers say:
A UMM can solve NP-complete problems in polynomial time. However, by virtue of its information overhead, a UMM needs only an amount of memory cells (memprocessors) that grows polynomially with the problem size.
This is a promising property of quantum computers, but the memcomputing prototype uses standard microelectronic technology so that it can be easily realized.
“Although the particular machine presented here is eventually limited by noise – and will thus require error-correcting codes to scale to an arbitrary number of memprocessors – it represents the first proof of concept of a machine capable of working with the collective state of interacting memory cells,” say the researchers.
Images from Fabio L. Traversa, Massimiliano Di Ventra, and Pixabay.