At present, the research and development direction of AI chip is mainly divided into two kinds: one is FPGA (field programmable gate array) and ASIC (application specific integrated circuit) chip based on traditional von Neumann architecture, and the other is brain like chip designed to imitate the structure of human brain neurons. FPGA and ASIC chips have formed a certain scale both in R & D and application; Although brain like chips are still in the early stage of research and development, they have great potential and may become the mainstream of the industry in the future.

The main difference between the two development routes is that the former follows the von Neumann architecture and the latter adopts the brain like architecture. Every computer you see uses the von Neumann architecture. Its core idea is that the processor and memory should be separated, so there is CPU (central processing unit) and memory. Brain like architecture, as its name implies, imitates the structure of human brain neurons, so CPU, memory and communication components are integrated together.

After that, due to the rapid development of high-definition video and game industry, GPU (graphics processor) chip has achieved rapid development. Because GPU has more logic operation units for processing data, it belongs to a high parallel structure, and has more advantages than CPU in processing graphics data and complex algorithms. In addition, because AI has many model parameters, large data scale and large amount of calculation for deep learning, GPU replaced CPU and became the mainstream of AI chips at that time.

FPGA can be understood as “universal chip”. Users define the connection between these gates and memory by burning the FPGA configuration file, and design the hardware circuit of FPGA with hardware description language (HDL). Each time the burning is completed, the hardware circuit inside the FPGA has a certain connection mode and has certain functions. The input data only needs to pass through each gate circuit in turn to get the output result.

FPGA has a strong development momentum and will have great prospects in the future

In the vernacular, “universal chip” is a chip that has what functions you need and what functions it can have. Although it is called “universal chip”, FPGA is not without defects. Because the structure of FPGA has high flexibility, the cost of a single chip in mass production is also higher than that of ASIC chip, and in terms of performance, the speed and energy consumption of FPGA chip are compromised compared with that of ASIC chip. In other words, “universal chip” is a “generalist”, but its performance is not as good as ASIC chip, and its price is higher than ASIC chip.

However, FPGA chips with reconfigurable characteristics have stronger adaptability when the chip demand has not yet become a scale and the deep learning algorithm needs continuous iterative improvement. Therefore, using FPGA to realize semi customized artificial intelligence chip is undoubtedly a safe choice. At present, the FPGA chip market is divided by American manufacturers Xilinx and Altera. According to the statistics of foreign media MarketWatch, the former accounts for 50% of the global market share and the latter accounts for about 35%. The two manufacturers occupy 85% of the market share and more than 6000 patents. There is no doubt that they are two mountains in the industry.

Before the large-scale rise of AI industrial applications, the use of general chips suitable for parallel computing such as FPGA can avoid the high investment and risk of developing customized chips such as ASIC. However, because of its low versatility, the high R & D cost of ASIC chips may also bring high risks. However, if market factors are considered, ASIC chips are actually the development trend of the industry.

Since we want to develop artificial intelligence chips, some experts return to the problem itself and begin to imitate the structure of the human brain. There are hundreds of billions of neurons in the human brain, and each neuron is connected with other neurons through thousands of synapses to form a super huge neuronal circuit, which transmits signals in a distributed and parallel way, which is equivalent to large-scale parallel computing, so the computing power is very strong. Another feature of the human brain is that not every part of the brain is working all the time, so the overall energy consumption is very low.

This kind of brain chip is different from the traditional von Neumann architecture. Its memory, CPU and communication components are completely integrated, taking the digital processor as neuron and memory as synapse. In addition, on the brain like chip, the information processing is completely carried out locally, and because the amount of data processed locally is not large, the bottleneck between the traditional computer memory and CPU no longer exists. At the same time, as long as neurons receive pulses from other neurons, these neurons will act at the same time, so neurons can communicate with each other conveniently and quickly.

However, one of the challenges in the development of brain like chips is to imitate the neural synapses in the human brain at the hardware level, in other words, to design perfect artificial synapses. In the existing brain like chips, the information transmission in neurons is usually simulated by applying voltage. However, the problem is that in most artificial synapses made of amorphous materials, there are infinite possibilities for ions to pass through, so it is difficult to predict which way ions go, resulting in the difference of current output of different neurons.

To solve this problem, this year, MIT’s research team made a kind of brain chip, in which the artificial synapses are made of silicon and germanium, and each synapse is about 25 nm. When voltage was applied to each synapse, all synapses showed almost the same ion flow, and the difference between synapses was about 4%. Compared with synapses made of amorphous materials, their properties are more consistent.

Even so, brain like chips are still quite far away from the human brain. After all, there are hundreds of billions of neurons in the human brain, and now there are only millions of neurons in the most advanced brain like chips, less than one tenth of the human brain. Therefore, the research of this kind of chip still has a long way to go before it becomes a mature technology that can be widely used on a large scale in the market, but in the long run, brain like chip may bring a revolution in computing system.

Leave a Reply

Your email address will not be published. Required fields are marked *