China currently lacks access to the most advanced semiconductor processes, a situation that might persist for a considerable time. However, it seems Chinese specialists might have discovered a method to make their chips, produced using older processes, as fast as Western chips made with newer technologies.
Wei Shaojun, vice-chairman of the China Semiconductor Industry Association, stated that China has developed an AI processor, fabricated using a 14-nanometer process, that allegedly matches the performance of Nvidia’s latest 4-nanometer chips. The full details remain undisclosed, as it is still an experimental chip rather than a commercial solution. Nevertheless, reports suggest the secret lies in combining the processor with DRAM using a new three-dimensional packaging technology.
If the claim proves accurate, using this technology allows the chip to deliver energy efficiency of 2 TFLOPS per watt with an overall performance of 120 TFLOPS while consuming just 60 watts. In terms of energy efficiency, it surpasses Nvidia’s solutions, although it lags in absolute performance. However, this can be compensated by increasing processor numbers, with the record energy efficiency allowing this without additional total power consumption.
Regrettably, there is no information yet on when the first chips utilizing this technology might reach the market. Recent developments have shown China’s resolve to innovate within constraints, suggesting an eagerness to challenge global leaders like Nvidia. If successfully scaled, these advancements could redefine China’s position in the global semiconductor landscape, potentially influencing industry trends towards efficiency-driven innovation.
