TSMC taps AI for energy-efficient chips as Qualcomm targets enterprise PCs
The post TSMC taps AI for energy-efficient chips as Qualcomm targets enterprise PCs appeared on BitcoinEthereumNews.com. TSMC rolled out a new AI-based chip design method on Wednesday in Silicon Valley, aiming to cut the power demands of AI chips by up to tenfold. The company, which manufactures chips for Nvidia and other tech giants, says this change is necessary as current systems burn too much electricity. At full load, Nvidia’s AI servers can use up to 1,200 watts, the same as keeping 1,000 U.S. homes running non-stop. That kind of energy drain isn’t sustainable, and TSMC is reportedly trying to fix it with smarter design. The approach revolves around building chips with smaller pieces called chiplets, each made with different tech, all packaged into one. But it’s not just about throwing pieces together. These new packages are being designed by AI software from firms like Cadence Design Systems and Synopsys, not by engineers alone. Cadence and Synopsys beat engineers on speed and accuracy Jim Chang, deputy director at TSMC’s 3DIC Methodology Group, showed off the results. Using Cadence and Synopsys software, chip designs that once took two days of human effort were finished by AI in five minutes. “That helps to max out TSMC technology’s capability, and we find this is very useful,” Jim said during his talk. The company sees this speed boost as key to getting more efficient chips to market faster. But not every problem can be solved with smarter code.Kaushik Veeraraghavan, an engineer at Meta’s infrastructure division, said during his keynote that the current chip manufacturing model is hitting physical walls.Moving data in and out of chips with traditional wires is slowing things down. Switching to optical connections could fix that, but right now, they’re still too unreliable for large data centers. “Really, this is not an engineering problem,” Kaushik said. “It’s a fundamental physical problem.” At the same event, Qualcomm launched…
BitcoinEthereumNews2025/09/25 11:46