Trend丨Intel's future path of neuromorphic computing
Focus: artificial intelligence, chips and other industries
Welcome all guests to follow and forward
This is an emerging direction in the field of artificial intelligence, which aims to simulate the structure and function of the human brain to achieve more intelligent machine learning and intelligent systems.
This field involves multiple disciplines such as neuroscience, computer science, cognitive psychology, etc., and aims to develop more intelligent, adaptive and learning artificial intelligence systems.
"Brain-computer interface" and neurotechnology equipment are two sub-sectors in the field of brain-like artificial intelligence, and the brain-computer interface has huge potential for future market development.
According to a McKinsey research report, the global annual market size of brain-computer interfaces may be between US$70 billion and US$200 billion (approximately RMB 510 billion to RMB 1.4 trillion) from 2030 to 2040.
Tianfeng Securities Research Report believes that the brain-computer interface industry has broad market prospects and focuses on grasping the possible scenarios. Minsheng Securities Research Report stated that as an emerging industry, brain-computer interface technology is mainly divided into hardware and software layers.
The hardware layer includes EEG acquisition equipment and external control equipment, and the software layer includes biological signal analysis, core algorithms, communication computing, and security and privacy. These technologies are all involved by Chinese companies.
Recently, Intel Corporation of the United States released a large-scale neuromorphic system called Hala Point, which aims to support cutting-edge research in the field of brain-like artificial intelligence and solve challenges in artificial intelligence in terms of efficiency and sustainability.
Looking back on Intel’s journey towards neuromorphic computing: In 2017, Intel Research launched a neuromorphic chip Loihi1, and in 2020 launched Pohoiki Springs based on Loihi 1.
In 2021, Intel Research launched the second-generation Loihi chip, and recently Intel also launched the large-scale neuromorphic system Hala Point.
Intel Hala Point's underlying Loihi chip
The main feature of using the Loihi chip is that it contains a smallest computing unit that simulates the structure and operation of neurons in the biological brain.
The Loihi chip contains multiple such units. Each smallest computing unit is composed of computing logic and corresponding storage, so it is an integrated storage and computing chip.
The Loihi 1 chip uses Intel's 14nm process. Pohoiki Springs, based on the Loihi 1 chip, released in 2020, is approximately 5U in scale and contains 768 Loihi1 chips.
There are nearly 100 million neurons in the Pohoiki Springs system. For comparison, the human brain has 86 billion neurons, which means that Pohoiki Springs is equivalent to 1/800 of the brain's neurons.
Intel's most advanced neuromorphic system
According to a press release from Intel, the newly released Hala Point is slightly larger than the original Pohoiki Springs, increasing from 5U to 6U. This increase is not large, but the number of neurons has increased 11 times, from 100 million to 1.15 billion.
In terms of size, this system is about the size of an oven or a suitcase, and its neurons are 1/80 of the size of the human brain.
The Hala Point system was initially deployed at Sandia National Laboratories in the United States and consists of 1,152 Intel Loihi 2 processors.
Compared with the Loihi generation, significant improvements have been achieved in density, computing power, speed, and interconnection characteristics.
Loihi 2 uses Intel 4 process technology, upgrading from Intel's 14nm process node to Intel 4. The process alone has significantly improved the transistor density and energy efficiency inside the chip.
1.15 billion artificial neurons and 128 billion artificial synapses distributed across 140,544 processing cores, with a maximum power consumption of only 2,600 watts.
When running traditional deep neural networks, it can perform 20 quadrillion operations per second.
When used in a bionic spiking neural network model, Hala Point is able to run all of its 1.15 billion neurons at real-time speeds 20 times faster than the human brain.
Especially when the number of neurons is low, its speed can be 200 times faster than the human brain. In other words, 80 Hala Points stacked together are equivalent to a neuromorphic computing cluster of the human brain scale.
Intel's future direction of neuromorphic computing
First, in terms of hardware, we will continue to promote the optimization and innovation of the architecture, and at the same time, iterate the process technology to achieve larger scale and better energy efficiency results.
This design directly benefits from the improvement of process nodes. Currently, Intel 4 process is used, and in the future there may be more advanced processes such as Intel 3 and Intel 18A, which will further promote the growth of neuron scale and is expected to double or even increase.
The second strategy is to polish the software better. Over time, since we started building the software stack in 2021, many application scenarios have changed, including the artificial intelligence framework.
From initially processing vision and perception applications to now needing to run larger-scale models, the demands are increasing, and continuous improvement of the software is required.
The third strategy is to continue to develop various applications in global cooperative communities such as INRC, and look forward to rapid large-scale applications in certain areas.
Looking back at the development of my country's brain-computer interface industry
Research and development related to brain-computer interfaces has been carried out in many fields such as bionics, medical diagnosis and intervention, and consumer electronics. From the perspective of product development, my country's representative brain-computer interface companies such as Qiangnao Technology, NaoLu Technology, and Rouling Technology have launched a series of consumer-grade brain-computer interface products, with application scopes covering children's education, entertainment, medical care, health and other fields.
According to estimates by the China Electronics Standardization Institute, the global brain-computer interface industry market size has exceeded US$1.5 billion, while the domestic brain-computer interface industry market size is only around RMB 1 billion, accounting for less than 10% of the global market size. The industry still has a lot of room for development.
At the same time, although domestic brain-computer interface technology is developing rapidly, my country has not yet established a relatively unified standard system in the field of brain-computer interface data, brain signal decoding algorithms, and signal acquisition equipment. This may lead to a lack of order in the market in the future and hinder the development of the brain-computer interface industry.
Currently, among my country's representative enterprises, most of them are developing non-invasive brain-computer interface products, and only a few companies such as Brain Tiger Technology and Boreikon Technology are conducting research on invasive brain-computer interface products.
In the field of non-invasive brain-computer interfaces, companies such as Qiangnao Technology and Huinao Intelligence have already launched relatively mature consumer-grade products.
In recent years, although a large number of start-ups have emerged in my country's brain-computer interface industry, and the number and amount of patent applications, financing events have increased significantly, from the application level, brain-computer interface products are still some distance away from large-scale industrialization.
In order to promote the advancement of brain-computer interface technology and realize the industrialization of brain-computer interface, the country has proposed, on the one hand, to strengthen basic research in brain science and promote the popularization of brain science. On the other hand, it encourages brain-computer interface companies to accelerate the application of products in typical fields such as medical treatment, rehabilitation, virtual reality, and education.
Conclusion : Progress of the Neuromorphic Ecosystem Project
The Intel Neuromorphic Research Community has launched eight Intel-supported university projects, including George Mason University, Queensland University of Technology, Graz University of Technology, University of Zurich, Brown University, Pennsylvania State University, University of Waterloo, and University of Göttingen.
Research projects include adaptive robot positioning, wireless biomimetic sensor pulse decoding for brain-computer interfaces, neuromorphic Bayesian optimization, auditory feature detection, and novel brain-like architectures and algorithms.
Since its founding in 2018, the Intel Neuromorphic Research Community has grown to more than 180 members, including university labs, government agencies, and leading global companies such as Accenture, Lenovo, Logitech, and Mercedes-Benz.
Going forward, Intel Labs will continue to provide developers with new tools to make it easier for them to develop applications that solve real-world problems and continue to support community research.
References: Semiconductor Industry Observer: Intel has made significant progress in neuromorphic computing; Qianzhan.com: 200 times faster than the human brain! Intel releases a large-scale neuromorphic system, and the brain-computer interface trillion-dollar track has great development potential; Zhiding.com: Intel releases a series of new progress to promote the application development of neuromorphic computing
The articles and pictures published on this public account are from the Internet and are for communication purposes only. If there is any infringement, please contact us and reply. We will deal with it within 24 hours after receiving the information.
Recommended reading:
京公网安备 11010802033920号