"Beyond CMOS" - Matrix computation engine using GPUs

Scientists have taken a common component of digital devices and endowed it with a previously unobserved capability, opening the door to a new generation of silicon-based electronic devices. In a research article published in July 2019 in Physical Review Letters, the scientists explain how they created a metal oxide—the "MO" in "CMOS"—equipped with an additional important function. Instead of simply being a passive element of the on-off switch in a CMOS transistor, the new metal oxide activates electrical current flow all by itself. The finding could one day help move computing into an era often called "beyond CMOS." The oxide material creates current in nearby pure, "undoped" silicon, the workhorse semiconductor of the electronics industry. The conductivity in silicon takes place in a very thin region just nine atomic layers thick. You'd need to stack 100,000 such layers equal to the width of a human hair. This capability—to induce current in silicon—marks a major step forward for a material that has previously been thought of as being of limited value; it has performed the on-off duties of an insulator very well but it hasn't been considered for the crucial current-creating capacity on which all transistors rely.

Read More: https://techxplore.com/news/2019-07-uncovers-capability-semiconductor-material.html

DistME (Distributed Matrix Engine) technology can analyze 100 times more data 14 times faster than the existing technologies. Also called CuboidMM, this method forms matrix multiplication in a 3-D hexahedron and then partitions and processes to multiple pieces called cuboids. The optimal size of the cuboid is flexibly determined depending on the characteristics of the matrices, i.e., the size, the dimension, and sparsity of matrix. The DistME technology developed by Professor Kim's team has increased processing speed by combining CuboidMM with GPU, which is 6.5 and 14 times faster than ScaLAPACK and SystemML respectively and can analyze 100 times larger matrix data than SystemML. It is expected to open new applicability of machine learning in various areas that need large-scale data processing including online shopping malls and SNS.

Read More:  https://techxplore.com/news/2019-07-distme-fast-elastic-matrix-gpus.html