AI is expected to reach the full potential of existing foundry capacity.
The problem of chip shortage has begun to hinder global economic growth and induce inflation, which cannot be solved in the short term.
The reasons for the chip shortage are complex, and the imbalance in the distribution of manufacturing capacity is an important part of it. Because the world has always paid close attention to the "high-end manufacturing nodes", the "mid-end nodes" such as 12 nanometers and 16 nanometers, which have accumulated rich history but are relatively inconspicuous, are in a state of surplus, while the traditional chip manufacturing nodes that have been eliminated by the times and have a lower technical level are basically out of the public eye. At present, Apple, Intel, Qualcomm and Nvidia are all promoting the production and construction of high-end chips, but the U.S. Department of Commerce said that 8%-10% of the total annual production capacity of global fabs (corresponding to 20 million to 25 million wafers) has been idle for a long time. If these old chips and foundry facilities can be upgraded to new usable capacity, will it become the key to overcoming the problem of chip shortage? Maybe there's a play.
Rebalancing capacity with demand
Intel CEO Pat Kissinger recently mentioned, "It is precisely because process nodes are difficult to complete the conversion efficiently and quickly, so a considerable part of the chip production capacity has not been effectively utilized." Using 'old nodes' will certainly not create 'new chips', so in order to alleviate the supply problem, we must explore a design route to achieve node conversion and upgrading." ”
However, this kind of node transformation, or "repositioning" plan, seems beautiful, but in practice, there are many problems. At present, engineering and technical resources have been very scarce, and limited manpower is busy designing new chips to help enterprises increase future revenue and profit margins. Moreover, repositioning is not much easier than designing a new chip, and who wants to take on such a laborious and unpleasant task? The answer may be AI – a technology that promises to rapidly update chip manufacturing facilities in an affordable way, making more powerful and lower-power chip products on the same production line. Synopsys, a manufacturer of AI chip design for many years, is eager to try it out.
Chip remake
Synopsys named the initiative "chip remake," and the basic idea is to use AI algorithms to automatically redesign chips for different nodes, compressing the work that previously took months or even years into just a few weeks, thereby helping customers save hundreds of millions of dollars in research and development costs.
Synopsys co-founder and co-CEO Aart de Geuss discussed the possibility in a keynote address at the annual ISSCC conference, saying that this approach is expected to bring an additional 25 million wafer capacity – equivalent to a year's current global foundry production , thus quickly alleviating supply pressures in a relatively short period of time. Dr. Aart de Geus emphasized, "In five years, there will be a major shift in the global chip supply chain. In this wave of efforts to increase production capacity, we believe that chip remake technology will become the key. ”
"Chip remake" can be understood as repairing old audio data in terms of process. Engineers can use new audio technologies to process old recordings to create new versions with higher sound quality. In addition, they were able to add new audio tracks, such as background music, thus re-producing new products. The cost of the entire process is only a fraction of the cost of re-recording. Cirque du Soleil's "Love" soundtrack is a typical case, and the same idea can also play a role in the chip field. Simply by introducing new process nodes, we can unlock new capacities in the old facility and make further optimizations.
From the perspective of practical results, chip remake has indeed brought a good solution, as shown in the figure below. It can be seen that the new chip has "higher performance", "smaller surface area" and "lower operating power consumption". What's more, the entire remake process takes only a few weeks plus an engineer, which is a farewell to the complicated planning that used to take months and require a team of designers to participate. This convenient upgrade method should be able to attract manufacturers such as Intel to convert old chip production lines to mid-range capacity such as Intel 16.
Figure: Chip remake, upgrading old equipment to new process nodes, and using AI technology to reduce chip power consumption and increase chip frequency.
-
This article is originally produced by Technology Walker, please do not reprint without permission