Memcomputer processes data in a similar way to the human brain

Researchers have built a prototype of a ‘memcomputer’, a computer that works by mimicking certain properties of the human brain. The findings could help develop computers that use ‘mem processors’.

Memprocessors consist of cooperating memory cells that take advantage of the ability to both store and process information. The machine presented in the paper is still very limited technologically and it is therefore also a proof-of-concept, the researchers write in the introduction. The machine was built using standard microelectronics so that theory could be easily converted into practice in a laboratory setting.

The authors of the study published a theoretical paper on a possible memcomputer earlier this year in which they showed that it should be easier for a memcomputing machine to solve “notoriously difficult computational problems, so-called NP-complete problems.” With the built prototype, the authors manage to solve an NP-full version of the subset sum problem in just one step, using a number of memprocessors that scale linearly with the size of the problem. The machine built by the researchers does suffer from noise, but it does offer possibilities for the future.

If a normal computer, or deterministic Turing machine, has to solve an NP-complete problem, such as the traveling salesman problem, the required processing power and memory can go up very quickly. In the traveling salesman problem, this is caused by an increase in the number of points through which it must pass. As a result, solving it can become almost impossible. Theoretically, this computer can solve problems in a short time. Other computers that could potentially solve such problems in the future include quantum computers.

Schematic of the memcomputer architecture used for this experiment to solve the subset sum problem Source: AAAS