Samsung released a black technology paper: replicating the brain with storage

A few days ago, the Samsung Electronics research and development team and Harvard University jointly published a research paper. They proposed a new method to “reverse engineer” (replicate) the human brain on a memory chip.

According to reports, the research paper was published in the scientific journal “Nature Electronics” (Nature Electronics), and the title of the paper is “copy-and-paste brain-based neuromorphic electronics”. The authors of the paper include Ham Don-hee, a researcher at the Samsung Advanced Institute of Technology and a professor at Harvard University, Park Hong-kun, a professor at Harvard University, Hwang Sung-woo, CEO of Samsung SDS, and Kim, vice chairman of Samsung Electronics. Ki-nam.

copy brain

In the paper, the researchers point out that nanoelectrode arrays developed by the two authors could be used to replicate the neural network connectivity map of the human brain. This connection diagram can then be copied into a high-density 3D network of solid-state memory chips.

Through this copy-and-paste technique, the authors hope to create a memory chip that can mimic the computing properties of the human brain, such as low Power consumption, rapid learning processes, environmental adaptability, automation and cognitive properties. This target technology has surpassed existing human scientific research achievements.

According to reports, the human brain includes countless neurons, and there are complex network connections between neurons, which realize the functions of the brain. Therefore, if you want to reverse engineer the human brain, you must first figure out the network connection diagram of neurons.

Born in the 1980s, neuromorphic engineering aims to mimic the structure and function of the human brain’s neural network on a semiconductor chip. But this is an extremely challenging technique, and until now, scientists have not figured out how many neurons are connected to each other to make up the complex functions of the human brain.

Faced with such complex challenges, the goal of neuromorphic engineering has since been adjusted. Instead of imitating the human brain through a chip, it develops related chips through revelation from brain function.

return to the original goal

However, the paper published by Samsung Electronics and Harvard University this time proposes another method that can return to the original goal of neuromorphism of brain reverse engineering.

It is reported that nano-electrodes can enter a large number of brain neurons, and can use their high sensitivity to record current signals. This massive intercellular parallel recording system can obtain information on neural network maps, discover the direction of interconnections between neurons, and demonstrate the strength of interconnections. From these recorded data, researchers can extract a graph of neural network connections.

The above network connection diagram can then be pasted into a network built by a memory chip. The memory chip can be the flash memory used in solid-state drives on the market, or a newer memory chip such as RRAM. Researchers can program memory chips so that the conductivity between each chip reflects the strength of the brain’s neuronal connections.

Quick copy

The paper goes a step further and proposes a strategy for quickly copying neural network connectivity graphs across memory chip networks. By directly connecting the electrical signals recorded between these cells, the memory chip network can learn and express the brain’s neural network connectivity map. In other words, researchers can directly download a map of the brain’s neural network connections and copy it to a memory chip.

It is estimated that the human brain has more than 100 billion neurons, and the number of so-called “synaptic connections” is more than 1,000 times that of neurons. Therefore, a memory chip capable of replicating the neural network diagram of the brain must be capable of storing 100 trillion neurons. Volume of virtual neuron and synaptic data.

Through the 3D memory integration technology, the above-mentioned huge number of memory chips can be integrated on a single chip. Samsung Electronics is currently the world’s leading manufacturer of 3D memory integration technology.

In a press release, Ham Don-hee, one of the authors of the paper, said: “The research vision we propose is very ambitious, and if we move toward this heroic goal, we will simultaneously advance machine intelligence, neural The frontier of science and semiconductor technology.”

The Links:   MCC26-14IO1B FS100R12KE3-B3