On September 23, a research team from Samsung Electronics, one of the world's largest semiconductor manufacturers, and Harvard University, a world-leading research institution, jointly published a research paper. In the paper, they proposed a new method to "copy and paste" neurons from the human brain onto a chip, creating a storage chip with unique computational characteristics close to those of the brain.
This paper proposes a groundbreaking method for replicating the neuronal connectivity map of the brain using nanoelectrode arrays and then pasting this map onto a high-density three-dimensional network of solid-state memory. Through this replication and pasting method, the authors envision creating a memory chip that approximates the brain's unique computational characteristics: low power consumption, convenient learning, environmental adaptation, and even autonomy and cognition—achievements currently unattainable with existing technologies.
The human brain comprises countless neurons connected by a complex network that enables the brain's functions. Therefore, to reverse engineer the human brain, it is essential to first understand the neural network connections.
This study proposes a method for reverting to the original neuromorphic targets of brain reverse engineering. Nanoelectrode arrays can efficiently access a large number of neurons, allowing for the recording of their electrical signals with high sensitivity. These massively parallel intracellular recordings provide information about the neuronal network connectivity map, indicating the location and strength of neuronal connections. Therefore, from these suggestive recordings, the neuronal network connectivity map can be extracted or “replicated.”
The copied neuron graph can then be "pasted" into a network of non-volatile memory, such as the commercial flash memory in a solid-state drive (SSD) that we use in our daily lives, or "new" memory, such as resistive random access memory (RRAM). Each memory is programmed to have a conductivity that represents the strength of the connections between each neuron in the copied graph.
This paper goes a step further, proposing a strategy for rapidly "pasting" neuronal wiring diagrams onto a memory network. A specially designed non-volatile memory network can learn and represent neuronal connectivity diagrams when directly driven by signals recorded within cells. This is a scheme to directly download the brain's neuronal connectivity diagrams onto a memory chip.
Given that the human brain is estimated to have around 100 billion neurons and 1,000 times more synaptic connections, the final neuromorphic chip will require approximately 100 trillion memories. Integrating such a massive amount of memory onto a single chip will be possible through three-dimensional memory integration, a technology led by Samsung that has ushered in a new era for the memory industry.
Leveraging its leading experience in chip manufacturing, Samsung plans to continue its research into neuromorphic engineering to expand its leading position in next-generation artificial intelligence semiconductors and drive the development of machine intelligence, neuroscience, and semiconductor technology.
