Tsinghua University has made meritorious contributions and successfully developed the "Taiji" photonic chip, which is 1,000 times faster than silicon chips in terms of computing speed, which is definitely a major breakthrough for Chinese chips.
Chips, also known as integrated circuits and microcircuits, are basically silicon-based chips, which interconnect the transistors, resistors, capacitors, inductors and other components and wiring required by the circuit, and make them on the semiconductor wafer of the silicon-based medium, and then package them.
This kind of chip has hundreds of millions, even billions, tens of billions of transistors, what are these transistors used for? After the power is on, the current can be interrupted, amplified, stabilized, and rectified.
With the development of Moore's Law, the chip has been iterated to 3nm, and with the development of AI, soon 3nm chips will not be enough.
What's even more embarrassing is that I don't have an advantage in the field of cutting-edge chips, and the basic material silica sand is monopolized by the United States, according to relevant data, the United States' monopoly on high-purity quartz sand has reached more than 90% of the global market share.
The most advanced EUV lithography machine is monopolized by ASML in the Netherlands, and the mass production of domestic lithography machines is only 90nm accuracy, and EUV photoresists and masks are basically monopolized by Japan.
In other words, the United States, Japan, and the Netherlands control the lifeblood of silicon-based chips, and we are very passive and difficult to make breakthroughs in high-end chips.
In addition, silicon-based chips perform operations by moving electrons, which consume electrical energy and also generate heat.
Take ChatGPT as an example, it can generate more than 500,000 kilowatt-hours of electricity in a day, and if it continues to be upgraded and popularized, the power consumed will be very huge.
Nvidia CEO Jensen Huang said, "In the end, AI will have to rely on photovoltaics and energy storage to carry the banner! Just thinking about how to improve computing power is not enough, and if it continues like this, computers alone will have to consume 14 earths of energy."
Altman, the CEO of OpenAI, also plans to raise $7 trillion in research and development to expand the production of AI chips worldwide.
This series of problems is more difficult for domestic chips, and the funds required are even larger. If you follow this path, not to mention whether you can take the lead in technology, you can't afford to spend money, and the energy consumed will make the global temperature rise.
Therefore, the future development must take other paths, and this time Tsinghua University has made meritorious contributions, and the "Taiji" photonic chip developed has successfully solved these problems.
Compared with silicon-based electronic chips, photonic chips have great advantages:
Faster speeds and more bandwidth
Photonic chips use optical signals to transmit, process, and convert data, and its speed is close to the speed of light, and at the same time, it has a huge bandwidth to transmit massive amounts of data, which is simply tailor-made for AI.
Lower energy consumption
Photons have basically no heat loss when transmitting data, and have a clear advantage over electronic chips in terms of energy efficiency.
Because the heat generated is very small, it can be densely packed and does not require a large number of heat dissipations when used.
Low latency
Photonic chips are extremely fast, and there is little to no delay when performing large, complex calculations.
Parallel processing
Photonic chips can transmit multiple frequencies of optical signals in the same fiber, allowing multiple data streams to be processed simultaneously and in parallel, making large-scale computing possible in AI applications.
At the same time, photonic chips can also reduce the dependence on the lithography process, and it does not require too advanced lithography machines to be successfully manufactured.
The team led by Dai Qionghai and Fang Lu of Tsinghua University has developed an optoelectronic chip that can be said to be of epoch-making significance.
Professor Dai Qionghai, currently an academician of the Chinese Academy of Engineering and dean of the School of Information Science and Technology of Tsinghua University, has done a lot of basic research work in stereo vision, three-dimensional reconstruction, and computing instruments, trying to reveal the laws of neurology and brain science, and provide support for the creation of a new generation of artificial intelligence.
Fang Lu is currently an associate professor and doctoral supervisor of electronic engineering at Tsinghua University, mainly engaged in the research of light field intelligent imaging theory and technology, proposed a new principle of unstructured light field perception, developed a scanning light field element imaging sensor, overcome the inherent contradiction between space and perspective, and realized billion-pixel dynamic light field imaging.
The team led by them abandoned the electronic chip computing method and turned to the development of optical computing architecture, and successfully developed the interfero-diffraction heterogeneous integrated chip Taichi, which realized 160TOPS/W general intelligent computing.
The H100 has a computing power of 2000TFLOPS and a power consumption of 700W, which is equivalent to 2.85TFLOPS/W.
In terms of energy consumption ratio, Tai Chi is 56 times that of H100.
You must know that H100 is a product that NVIDIA has iterated for more than 20 years, if the Taiji light chip has been iterated for more than 20 years, what will be the effect?
In November 2023, the team's paper on the optoelectronic fusion computing framework published in the journal Nature showed that the first fully analog optoelectronic intelligent chip developed by the team has a computing speed of 4.6 petaflops per second, which is 15 times that of NVIDIA A100.
The energy consumption is more than 4 million times that of high-performance chips, which means that the power of high-performance electronic chips running for one hour can support the operation of optoelectronic chips for more than 500 years.
In the field of traditional chips, we have always been "stuck" and suppressed, but in the field of optical chips, we have achieved the world's leading.
This reminds me of our automobile industry, in the field of traditional fuel vehicles, we have a big gap with foreign giants, Toyota, Volkswagen, Ford, General Motors, Kia and other overseas brands are selling better than us.
Domestic car companies have cooperated with overseas brands, giving birth to Shanghai Volkswagen, FAW Toyota, BMW Brilliance, Changan Ford, etc., although the sales have risen, but we have not obtained the core technology, and the profits have also been taken away by overseas car companies.
With the birth of new energy vehicles, we have made great efforts to develop, and finally mastered the core technologies such as "three electrics" and intelligent driving, improved the relevant industrial chain, trained a large number of skilled workers, and finally became the leader in the field of new energy vehicles.
Tesla in the field of new energy vehicles has also begun to lay off employees, and Musk even said that nine of the world's top ten electric vehicle companies will be Chinese companies in the future.
Nowadays, we have also opened up a new front in the field of chips, and are in a leading position, if we strive to engage in research and development, while actively supporting the industrial chain, training technical talents, then in the future we will also fight a turnaround in the field of chips.
Using the high performance, high bandwidth and low power consumption of optical chips, we will develop our own AI chips, and then feed back the chip research and development through AI technology, and even all walks of life, when China's science and technology will dance on the world stage.
I'm a science and technology master, if you like it, give it a like!