Can chips copy brain ideas?The next technological revolution may be detonated by the “core”

Recently, the top international journal “Nature Electronics” (Nature Electronics) published a cutting-edge academic paper jointly authored by Samsung’s top engineers and Harvard University researchers, entitled “Neuromorphic Electronics Based on Copy and Paste Brain”. The paper puts forward the vision of developing a simulated human brain chip. By duplicating brain neurons, the Electronic chip can achieve unique computing characteristics similar to the human brain, bringing low power consumption, easy learning, adapting to the environment, etc., and even having a certain degree of autonomy Sex and cognition. This scientific and technological concept once again demonstrates Samsung’s advanced strength in the field of technology research and development such as semiconductor chips.

Samsung’s forward-looking technology research and development is the industry’s sustainable power

Today, when artificial intelligence technology is widely used, the research and development of forward-looking technology around it has become more and more in-depth, and the research on brain-like chips has received more attention.

As an important computing organ for human beings, the brain has huge computing power, with at least 100 billion neurons, which is approximately equal to the number of stars in the Milky Way. These neurons constitute 10-15 neural connections, connecting complex connected nerves end to end, with a total length of more than 180,000 kilometers. And having super learning ability and low power consumption is also the goal that the brain-like chip research field has been pursuing, but the process of realization is extremely challenging. Because scientists have so far known little about how neurons are connected to each other to construct the complex functions of the human brain. But this does not prevent scientific and technological workers from using the brain as an “inspiration” to develop electronic chips with the structure and functions of the brain’s neural network.

In the paper jointly published by Samsung and Harvard this time, its research goal is mentioned-the neuromorphology of brain reverse engineering, which provides the possibility for the research and development of brain-like chips. The paper pointed out that the nanoelectrode array can effectively enter the brain neurons, and record current signals with extremely high sensitivity, and then grasp the location of the interconnection of neurons and the strength of the interconnection. From these records, R&D personnel can extract or “copy” neuron connection diagrams and “paste” them to non-volatile memories (Non-Volatile Memories), such as commercial flash memory in daily solid-state drives (SSD) , Or variable resistance memory (RRAM) and other new types of memory, etc., and then through the programming of each memory, it can make use of conductivity to reproduce the connection strength of the neurons in the copied connection diagram.

Samsung’s top engineers and Harvard University researchers put forward the vision of developing a simulated human brain chip

The paper also pointed out that the human brain has at least 100 billion neurons, and the synaptic connections they form are more than 1,000 times the number of neurons. Therefore, the neuromorphic chip ultimately used for “replication” will have the capacity to store 100 trillion virtual neurons and synaptic data.

In this regard, Samsung Electronics stated that the “3D integration technology” led by Samsung has created a new era in the memory industry, which can make it possible to integrate so many memory chips on one chip. In the future, using its leading experience in the field of semiconductor manufacturing, Samsung will continue to conduct neuroengineering research to expand its leading position in the field of next-generation artificial intelligence semiconductors.

Through the papers, people have been enough to intuitively feel the research and development strength of Samsung’s brain-like chips. And on the road to the future of artificial intelligence, Samsung has done more to explore.

For example, Samsung has successfully developed the industry’s first high-bandwidth memory HBM-PIM that integrates artificial intelligence processing capabilities. A DRAM optimized AI engine is added to each storage bank, which can directly provide the processing power required by the data storage location, and can achieve parallel processing and greatly reduce data movement. This breaks through the von Neumann bottleneck discussed in the industry, and can accelerate the processing speed of large-scale data operations in data centers, high-performance computing systems, and AI mobile applications, bringing more than twice the system performance and reducing consumption by nearly 70%. Power.

At the same time, at the ISSCC (International Solid-State Circuits Conference, the highest-level conference in the field of integrated circuit design recognized by the world’s academic and business circles), known as the “Integrated Circuit Olympiad”, the number of accepted papers by Samsung also fully reflects its status. Outstanding achievements in this field. At the ISSCC 2020 conference, Samsung has a total of 13 papers selected. At this year’s ISSCC, Samsung has won many papers in many fields, including image sensors, 3nm, memory, AI chips, and leading 5nm mobile phone SoCs and data converters.

Thanks to the accumulation of research and development in the field of scientific research, Samsung has made outstanding contributions in promoting the industry’s forward-looking technology towards commercial production and application. For example, it has been reflected in aspects such as foundry and CIS research and development that have attracted much attention from the industry in recent years.

From foundry to CIS R&D, look at Samsung’s fast pace

As one of the leaders in foundry companies, Samsung has transformed innovation into technology that promotes industry progress, constantly iterating wafer production processes to meet market demand for high-performance chips, and breaking the limits considered by industry experts .

At the Samsung Foundry Forum held recently, Samsung announced that the company’s 3nm generation process GAE will be put into mass production in 2022, 3nm generation 3GAP will be mass production in 2023, 2nm generation 2GAP process will be available in 2025 . Especially at 3nm, Samsung will implement the Gate-All-Around transistor technology that is unanimously optimistic in the industry. This technology can continue the existing semiconductor technology route, enhance the gate control capability, and overcome the physical scaling and performance limitations of the current technology. It will provide strong technical support for the development of data centers and artificial intelligence in the future.

In terms of CIS (contact image sensor) research and development, Samsung has performed equally well. When the innovation of smart phones has gradually entered a bottleneck, mobile phone manufacturers have set their sights on the field of mobile phone cameras. This has become the driving force for various CIS suppliers to continuously upgrade their technology. However, when competitors in the industry were deeply involved in cameras with several thousand pixels, Samsung took the lead in advancing CIS to the “100 million” level.

In September of this year, Samsung officially launched its first 200-megapixel (200Mp) resolution image sensor ISOCELL HP1 based on 0.64m pixels. This product not only has ultra-high resolution, but also has a compact package that can be well integrated into All kinds of handheld devices today. With the ISOCELL HP1 blessing, even if you take pictures indoors or in the evening and other low-light environments, you can take high-resolution photos and see clearer details. What’s more surprising is that ISOCELL HP1 can also shoot 8K video at 30 frames per second (fps) and minimize field loss.

Since entering the semiconductor industry in 1974, Samsung has created miracles in a number of semiconductor fields, and is also making a broad layout for the future technology industry. As Donhee Ham, a researcher at Samsung Electronics Advanced Technology Research Institute, said, “Our vision is very ambitious, but working towards such a grand goal, we will push the boundaries of machine intelligence, neuroscience and semiconductor technology.”