Home > Article > Technology peripherals > When will the AI chip shortage ease? TSMC: CoWoS production capacity shortage will be resolved in one and a half years!
News on September 7, according to Taiwanese media "Economic Daily" reported that TSMC Chairman Liu Deyin said yesterday that the current shortage of AI chips is mainly due to insufficient CoWoS advanced packaging production capacity. TSMC is doing its best to support customers. It is expected that in one and a half years Technical capacity can catch up with customer demand, and the current shortage should be a temporary, short-term phenomenon. In addition, the development of semiconductor technology "has reached the exit of the tunnel. There are more possibilities outside the tunnel, and we are no longer bound by the tunnel."
"SEMICON Taiwan 2023 International Semiconductor Exhibition" opened on September 6. Liu Deyin attended the "Master Forum Special Speech" and gave a speech on the topic of "Semiconductor Technology in the Artificial Intelligence Era" and accepted an interview, releasing the above information.
Liu Deyin pointed out that AI has evolved to a new level. In addition to facial recognition, translation or product recommendation in the past, it can now write poetry, create, write reports and programs, and can even design Internet circuits that are comparable to humans. ) and become an assistant in human life.
Liu Deyin believes that the amazing breakthroughs in AI applications are due to three factors, including the innovation of efficient deep learning algorithms, the availability of large amounts of training data on the Internet, and the innovation of semiconductor technology to achieve efficient and energy-saving computing.
Liu Deyin said that the computing and memory required for AI are still developing rapidly. In the past, semiconductors were dedicated to shrinking, but now semiconductor technology has shifted to 3D design. In addition to HBM3 high-bandwidth memory, 2.5D and 3D SOIC are increasingly important for AI. Successful Examples include Nvidia's GA100 and GH100 chips, which can pack 54 billion and 80 billion transistors respectively, as well as the AMD MI300 accelerated processor, which can pack 146 billion transistors, and Cerebras' wafer-level AI processing, which can pack 2.6 trillion transistors. Device WSE-2 etc.
Liu Deyin said that the development of semiconductor technology in the past 50 years has been like walking in a tunnel (commonly known as the limit of Moore's Law in the industry). There is a clear road ahead. Now we have reached the exit of the tunnel, and there are more possibilities outside the tunnel. The so-called clear path means that everyone knows that transistors must be shrunk and the number of transistors must be increased. Now this tunnel feels like it is about to be walked out, but it has entered another more complex and challenging tunnel brought about by 3D IC advanced packaging. , will bring more unlimited imagination space and development possibilities to the semiconductor industry.
Editor: Xinzhixun-Linzi
The above is the detailed content of When will the AI chip shortage ease? TSMC: CoWoS production capacity shortage will be resolved in one and a half years!. For more information, please follow other related articles on the PHP Chinese website!