[ad_1]
Researchers from Lund College have created a nanowire-based gadget, which may permit computer systems to carry out processing in-memory — fixing a long-standing bottleneck and doubtlessly boosting the efficiency and vitality effectivity of computer systems significantly.
“Processors have developed rather a lot over a few years,” says doctoral scholar and co-author Saketh Ram Mamidala, in an understated abstract of break-neck progress, which has taken us from the four-bit 750kHz Intel 4004 to at this time’s multi-core many-gigahertz beasts in simply fifty years. “On the reminiscence aspect, storage capability has steadily elevated, however issues have been fairly quiet on the perform aspect.”
A 3D-stacked nanowire-based transistor-and-RRAM pillar might show key to boosting computing efficiency. (📷: Saketh Ram et al)
That is the place the group targeted its efforts: Constructing 3D circuits which combine each reminiscence and processing capabilities, taking resistive RAM (RRAM) and including new features — together with, unusually, what Mamidala says is the power to work “with out a energy provide.”
“Our model [of RRAM] is a nanowire with a transistor on the backside, and a really small reminiscence ingredient positioned additional up on the identical wire,” says Lars-Erik Wernesson, professor of nanoelectronics and co-author of the paper. “This makes it right into a compact built-in perform the place the transistor controls the reminiscence ingredient.”
“The thought has been round earlier than, however it has confirmed tough to attain efficiency. Now, nonetheless, we’ve proven that this may be achieved and that it really works surprisingly properly.”
The group’s know-how is significantly quicker than conventional CMOS equivalents, whereas taking over a much smaller footprint. (📷: Saketh Ram et al)
Whereas it is early days for the group’s know-how, it is potential it may assist resolve the von Neumann bottleneck in computing: The truth that it takes time to switch knowledge between the processor and the reminiscence, making a bottleneck which saps the general efficiency for all however the smallest of information.
By processing in-memory, and with a dramatic discount in latency, the group’s gadget may provide large positive factors — notably for compute- and memory-hungry workloads like machine studying.
The group’s work, which resulted within the creation of single-transistor and single-memory gadgets able to Boolean logic operations, has been revealed within the journal Nature Electronics underneath closed-access phrases.
[ad_2]
