Researchers from University of Michigan devised a component on a chip capable of reducing energy consumption in computers.
Use of semiconductors in computer chips provide processors and memories that are fast in processing large amount of data. However, these semiconductor chips are not efficient as they have to wait for data input and output. Now researchers from University of Michigan developed memristors—a combination of memory and resistor—programmed to have different resistance states. This ability of memristors enables it to store information as resistance levels. Furthermore, it enables memory and processing in the same device that reduces data transfer. Therefore, memristors provide an efficient alternative to conventional method that stores the memory separately from the processor. However, memristors can have resistances that are on a continuum, which restricts its application in numerical calculations as it does not use ordinary bits— 1 or 0. Memristors can be an advantage for computing that mimics the Neuromorphic machine language as it benefits from its analog nature.
The researchers overcame the problem with numerical calculations by digitizing the current outputs. It was observed that the memristors were able to map large mathematical problems into smaller blocks within the array. Such effective problem solving ability improves the efficiency and flexibility of the system. Computers laced with memristors offer wide variety of applications in machine learning and artificial intelligence algorithms. Furthermore, various tasks based on matrix operations such as simulations used for weather prediction can be easily processed by such computers. Operations that multiply and sum the rows and columns can be processed simultaneously, once the memristors are set to represent numbers with a set of voltage pulses along the rows. The answers to such processes is stored in the current measured at the end of each column. Conventional processor perform the same task by reading the value from each cell of the matrix to perform multiplication and then sum up each column in series. The research was published in the journal Nature on July 13, 2018.