On September 25th, Global Foundries launched the 12LP+ process at its global technology conference, focusing on artificial intelligence training and reasoning applications. Compared to the previous generation 12LP process, the 12LP+ offers a 20% performance increase or a 40% power reduction with a 15% reduction in logic area.
In addition, a key feature of the 12LP+ process is the high-speed, low-power 0.5V SRAM memory cell that supports fast, energy-efficient data transfer between the processor and memory, which is the labor of the computing and wireline infrastructure market. Important requirements for smart applications.
Global Foundries provides design reference packages for artificial intelligence applications and design/technical joint development (DTCO) services for the 12LP+ process, both of which allow customers to view artificial intelligence circuit designs from a holistic perspective to reduce energy consumption and cost.
In addition, 12LP+ also has a new silicon interposer for 2.5D packaging, which helps integrate high-bandwidth memory with the processor for fast, energy-efficient data processing.
The 12LP+ solution uses ARM's Artisan® physical IP for Global Foundries and POPTMIP for artificial intelligence applications. Both solutions will also be applied to Global Foundries' first generation 12LP platform.
Gus Yeung, general manager and researcher at the Physical Design Group, said: "Artificial intelligence, automotive and high-end consumer mobile are just a few of the growing applications for meeting the urgent needs of high-performance SoCs. With the support of ARM Artisan physical IP and advanced processor design, Global Foundries' 12LP+ will help designers design products faster and more efficiently."
Michael Mendicino, vice president of digital technology solutions at Global Foundries, said: "The launch of 12LP+ is the result of Global Foundries' strategy of providing differentiated solutions to its customers. Global Foundries is able to scale designs without disrupting workflows compared to other solutions. Scale, which is very cost effective."
Michael Mendicino further stated: "For example, as an advanced 12nm technology, our 12LP+ solution has provided customers with most of the performance and power advantages they want from the 7nm process, but their NRE (non-repetitive engineering) The average cost is only about half, which brings significant cost reduction. In addition, because the 12nm node runs longer and matures, customers will be able to quickly streamline to fully grasp the growing demand for artificial intelligence technology. ”
According to Global Foundries, 12LP+PDK is now available and the company is currently working with several customers. It is expected to be released in the second half of 2020 and mass production in Fab 8 in Malta, New York in 2021.