Samsung Electronics and researchers from Harvard University presented a future vision for neuromorphic chips, a next-generation artificial intelligence semiconductor technology.

This paper, authored by Don Hee Ham, Fellow at Samsung Electronics Advanced Institute of Technology and Professor at Harvard University; Hong Keun Park, Professor at Harvard University; Sung Woo Hwang, President of Samsung SDS; and Ki Nam Kim, Vice Chairman of Samsung Electronics, was published in the world-renowned academic journal 'Nature Electronics' on the 23rd (local time in the UK).

This paper proposes a technological vision for a neuromorphic chip that reproduces the brain's unique functions by measuring the electrical signals of neurons in a brain neural network with ultra-high sensitivity using nanoelectrodes, 'copying' the connection map between neurons, and 'paste' the copied map onto a memory semiconductor.

Copying neural network maps through ultra-high-sensitivity measurements is achieved via an array of nanoelectrodes that penetrate neurons. By penetrating the neurons, measurement sensitivity is enhanced, allowing the detection of minute electrical signals generated at the neuronal interfaces. Consequently, these interfaces can be located to map the neural network. This is a technology that Samsung Electronics has been continuously collaborating with a Harvard University research team since 2019.

In particular, Samsung Electronics proposed a completely new concept of neuromorphic semiconductor in which a copied neural network map is pasted onto a memory semiconductor, allowing each memory to act as a point of contact between neurons.

In addition, while it takes a significant amount of time to construct a neural network map by analyzing a vast amount of signals measured in a neural network using a computer, a groundbreaking technical perspective was presented to rapidly download the neural network map by directly driving a memory platform with the measured signals.

This platform can utilize commonly used memory such as flash and other forms of non-volatile memory such as resistive memory (RRAM).

Meanwhile, to ultimately implement the approximately 100 trillion neuron contacts in the human brain into a memory network, memory density must be maximized. To this end, the use of cutting-edge semiconductor technologies was proposed, such as 3D flash stacking technology and 3D packaging using TSV (Through-Silicon Via) applied to high-performance DRAM.

This study is also significant in that it presented a vision for next-generation artificial intelligence semiconductors by combining neuroscience and memory technology with the participation of technology leaders from academia and industry.

Donhee Ham, a Fellow at Samsung Electronics’ Advanced Institute of Technology, stated, “The bold approach proposed in this paper will help expand the boundaries of memory and system semiconductor technology and further advance neuromorphic technology.”

Samsung Electronics plans to secure technological leadership in the field of next-generation artificial intelligence semiconductors by continuously focusing on neuromorphic research based on its existing semiconductor technology capabilities.