Samsung Electronics, a global leader in innovative semiconductor technology, announced today a breakthrough that brings the world one step closer to neuromorphic circuits that better resemble the brain. The idea was conceived by top engineers and academics from Samsung and Harvard University, and it was published in Nature Electronics as a Perspective report named ‘Neuromorphic electronics based on copying and pasting the brain.’ The co-corresponding authors are Donhee Ham, a Fellow of the Samsung Advanced Institute of Technology (SAIT) with a team of Harvard University Professor, Hongkun Park, a Harvard University Professor, Sungwoo Hwang, Ceo and President of Samsung SDS as well as former Head of SAIT, along with Kinam Kim, Vice Chairman, and Chief executive officer of Samsung Electronics.
The core of the writers’ vision may be summed up in two words: ‘copy’ and ‘paste.’ The study proposes utilizing a groundbreaking nanoelectrode array created by Dr. Ham and Dr. Park to duplicate the brain’s neural connection map and paste it out on a high-density three-dimensional matrix of solid-state memory, a technology in which Samsung has been a global leader. The authors want to develop a memory chip that mimics the brain’s unique computing characteristics, such as low power, easy learning, adaptability to the environment, and perhaps even autonomy and cognition, which are now beyond the grasp of current technology.
The wiring map of the brain, which is made up of a vast number of neurons, is accountable for the brain’s activities. As a result, understanding the map is crucial for reverse-engineering the human brain. While the aim of neuromorphic engineering, which began in the 1980s, was to replicate the structure and function of neural networks on something like a silicon chip, it proved challenging since little is understood about how the brain’s higher functions are created by wiring together a vast number of neurons. As a result, rather than strictly replicating the brain, the aim for neuromorphic engineering has indeed been lowered to creating a device that is “inspired” by it.
This research offers a method to restore to the brain reverse engineering’s initial neuromorphic aim. The nanoelectrode array may successfully penetrate a significant number of hidden neurons, allowing for the high-sensitivity recording of their electrical impulses. The neural wiring map is based on these parallel computing intracellular recordings, which show where neurons link to each other and how strong those connections are. The neural wiring diagram may therefore be retrieved, or ‘copied,’ from these telltale recordings. By programming each memory’s conductance to represent the resilience of every neuronal connection in the copied map, the copied neuronal map could be ‘pasted’ to such a network of quasi memories – like commercial flash memories used in solid-state drives (SSD) as well as ‘new’ remembrances like resistive random access memories (RRAM).
The study goes even further, recommending a method for quickly pasting the neural wiring diagram onto such a memory network. When directly driven by intracellular recorded signals, a network of specifically designed non-volatile memory can learn and represent the neural connection map. This is a method of directly downloading the neural connection map of the brain onto a memory chip. The ultimate neuromorphic device would need 100 trillion or more memories, owing to the human brain’s additional $ 100 billion so about neurons including a thousand or thereabouts lot more synaptic connections. 3D memory integration, a technique pioneered by Samsung and which has ushered in a new age for the memory business, would allow such a large number of memories to be combined on a single chip.
Samsung plans to continue its work in neuromorphic engineering, using its leading expertise in chip production, to expand its leadership in the area of next-generation AI semiconductors. “The vision we offer is very ambitious,” Dr. Ham said. “However, pursuing such a noble aim will test the limits of machine intelligence, neurology, and semiconductor technology.”
The researchers proposed that the brain’s neural connection map may be reproduced using such a nanoelectrode array (the electric conductor which conducts electricity). The electrical impulses generated by a vast number of neurons may be recorded using this nanoelectrode array. According to the researchers, this would then inform the brain’s neural map where neurons link to each other and whether strong those connections are or not. The data might then be copied and pasted into a 3-dimensional matrix of solid-state memories. This capability may be used as a “shortcut” to developing AI systems that act like human brains, including the capacity to acquire new ideas, adapt to changing circumstances, and learn basic tasks.
The study also indicates that simply downloading the neural map onto such a computer chip may be a quicker method of pasting it. The idea of fully cognizant robots is both fascinating and alarming, especially in light of the extremes portrayed in numerous Hollywood films, novels, and television programs.
Research paper: Ham, D., Park, H., Hwang, S. et al. Neuromorphic electronics based on copying and pasting the brain. Nat Electron 4, 635–644 (2021). https://doi.org/10.1038/s41928-021-00646-1