Neuromorphic computing, one that seeks to design algorithms and integrated circuits imitating the physics of the human brain and nervous system, is one of the branches with the most potential. In recent years, we have seen brain-computer interfaces and the singularity, with claims that our minds could be linked to cyberspace.
Now, Samsung Electronics has shared a new insight that takes the world a step closer to realizing neuromorphic chips that can better mimic the brain – by “copying and pasting” the brain’s neuronal connection map onto the chip.
The leading engineers and scholars from Samsung and Harvard University suggested a way to copy the brain’s neuronal connection map using a nanoelectrode array and to paste this map onto a high-density three-dimensional network of solid-state memories. Using this ‘copy and paste’ approach, the team envisions to create a memory chip that approximates the unique computing traits of the brain – low power, facile learning, adaptation to the environment, and even autonomy and cognition – that have been beyond the reach of current technology.
The original goal of neuromorphic engineering, launched in the 1980s, was to mimic the structure and functions of neuronal networks on a silicon chip. Given its difficulty (even until now, it is not known how a large number of neurons are wired together to create the higher functions of the brain), the goal has become to design a chip inspired by the brain rather than rigorously mimicking it.
Samsung’s approach involves returning to the original neuromorphic goal of the brain reverse engineering. It uses a nanoelectrode array that can effectively enter a large number of neurons so it can record their electrical signals with high sensitivity. These massively parallel intracellular recordings are then used to compile the neuronal wiring map by detailing the strength of different neural connections. This allows you to “copy” the neuronal wiring map, which can then be “pasted” into a network of non-volatile memories, such as a solid-state drive (SSD), or into resistive random access memories (RRAM). Each memory would be programmed so that its conductance corresponding to the strength of each neuronal connection in the copied map.
Obviously, years (perhaps decades) will pass for a first practical demonstration of the research. The human brain has over 100 billion neurons and thousands of times more synaptic connections. So, the ultimate neuromorphic chip will require 100 trillion or so memories. The study also indicates a quick way to paste the neural map into the memory network by downloading it directly.
Among the many challenges that the obtaining of a neuromorphic chip still poses is that of integrating such a large number of memories on a single chip, something that would be possible through 3D memory integration, a technology in which Samsung claims to be a leader and that leaves clearer their role in all this.