Samsung Electronics proposes next-generation memory semiconductor technology to ‘copy-and-paste’ brain neural networks

Researchers at Samsung Electronics and Harvard University in the United States presented a vision for the next-generation neuromorphic semiconductor chip in a method of ‘copy’ and ‘paste’ the brain neural network as it is. Neuromorphic semiconductors are semiconductors that mimic the neural networks of animals such as humans, and their goal is to reproduce high-level functions of the brain, such as cognition and reasoning.

Samsung Electronics is a thesis ‘Neuromorphic Electronic Technology Based on Brain Copy and Paste,’ written by Don-hee Ham, a fellow at Samsung Electronics Advanced Institute of Technology and professor at Harvard University, Hong-geun Park, professor at Harvard University, Seong-woo Hwang, president of Samsung SDS, and Ki-nam Kim, vice chairman of Samsung Electronics. electronics based on copying and pasting the brain)’ was published on the 23rd (UK local time) in the world-renowned scientific journal ‘Nature Electronics’. It is attracting attention for the fact that it combines neuroscience and memory technology with the participation of technology leaders from academia and industry and presented a vision for next-generation AI semiconductors.

This paper contains the vision of a neuromorphic chip technology that reproduces the brain’s unique functions by measuring the electrical signals of neurons (neurons) in the brain neural network, copying the connection map between neurons, and pasting the copied map into a memory semiconductor.

First of all, neural network map radiation is achieved through ultra-sensitive measurement that directly penetrates nano-electrodes into neurons. In the past, it was difficult to penetrate electrodes into neurons, but with the recent advancement of nanotechnology, it is now possible to arrange nanoelectrodes inside microscopic neurons. At the same time, it is possible to read insignificant electrical signals generated at the contact points (synapses) between neurons. Through this, it is a method of mapping the brain neural network by finding the contact points between neurons. Samsung Electronics and Harvard University have been collaborating and researching on this technology since 2019. The goal of this paper is to present a completely new concept of neuromorphic semiconductor, in which each memory acts as a contact point between neurons by pasting the copied neural network map into a memory semiconductor.

It also proposed a groundbreaking technology perspective that directly drives the memory platform with the measurement signal obtained through radiation. It is possible to significantly reduce the time to construct a neural network map compared to the method of analyzing a large amount of signals measured in a neural network with a computer. This platform can be utilized for commonly used memory such as flash and other types of nonvolatile memory, such as resistive memory (RRAM).

Ultimately, to realize about 100 trillion neurons in the human brain as a memory network, the memory density must be maximized. To this end, Samsung Electronics proposed the use of cutting-edge semiconductor technologies such as three-dimensional (3D) flash stacking technology and 3D packaging through silicon through-electrode (TSV) applied to high-performance DRAM.

Donhee Ham Samsung Electronics Fellow

Donhee Ham, a fellow at Samsung Electronics Advanced Institute of Technology, said, “The bold approach proposed in this paper will help to broaden the boundaries of memory and system semiconductor technologies and further develop neuromorphic technologies.”

By Kwon Dong-jun, staff reporter [email protected]

.

Leave a Reply

Your email address will not be published. Required fields are marked *