Incidentally, this isn’t the only improvement to EUV coming. To get below that limit, engineers had to make some major changes to the system’s optics and fix all the knock-on problems that followed. As engineers at ASML explained in Spectrum‘s August issue, the version of the technology in use now is limited to making patterns with a resolution of about the wavelength of its light, 13.5 nanometers. Even today’s advanced methods are slow to react to changes in chip temperature when compared to a thermal transistor, which can switch heat conductance on and off as quickly as 1 megahertz.Įxtreme ultraviolet lithography was decades in the making, only entering regular service a few years ago to print nanometer-scale patterns on the most advanced chips. Heat removal has been a long-standing limit on processor performance, and it’s only going to get more difficult as processors become collections of 3D-stacked chiplets. In November, researchers from University of California, Los Angeles reported the invention of a thermal transistor, the first solid-state device that uses an electronic signal to control the flow of heat. Thermal Transistors Can Handle Heat With No Moving Parts The hope is to steer talented students away from the allure of AI and other hot fields and steer them toward making the chips that make AI happen.ģ. CHIPS and Science Act set to pump tens of billions of dollars into chip manufacturing in the United States, the question has come up: “Who is gonna work in these new fabs?” As long-time contributor Prachi Patel reports, universities across the United States, especially those near fab construction projects, are revamping their semiconductor education offerings in response. Universities Are Building a Semiconductor Workforce As recently as August 2023, Jeff Dean, chief scientist at Google DeepMind, said that compared to other methods the TPU team had available, 26 of the TPU’s 37 blocks had better performing layouts because of the AI, and 7 of 37 performed equally well.Ģ. It has since led to an editorial expression of concern from Nature, and Kahng has retracted the editorial that originally accompanied Google’s paper. Kahng’s research largely backed the rival group’s take on things. When a version was leaked ahead of a major conference, things got ugly.Ī year later, a group led by IEEE Fellow Andrew Kahng reported research he said was meant to get the community past the unpleasant episode. A rival group at Google responded that it didn’t, but the company wouldn’t publish the rival’s results. In research published in Nature in 2021, Google claimed it beat top academic algorithms and human chip designers at finding the optimal layouts. At issue was a reinforcement-learning AI system that Google uses to do a key step in laying out chunks of logic and memory for its AI accelerator chips, or TPUs. In 2022, a schism in the chip design world and at Google erupted into the open. We’re already giving you a taste of the next year in semiconductors, and we can’t wait to see what you read most in 2024. You also wanted to know how semiconductors will play a part in combating climate change. You wanted to know what the future of Moore’s Law is and who would be making it happen. But you’re also a discerning and forward-looking person. For instance, you liked scandal and speed. If you were a reader of IEEE Spectrum in 2023 with an interest in semiconductors, this list of the top stories says some interesting things about you.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |