It’s never been a big deal that memory and processing are two separate pieces of hardware. With the advent of artificial intelligence and its ramifications, machine learning, and deep neural networks, moving information back and forth between “processing” and “memory” requires more time and energy that can be avoided. became.
Scientists are investigating ways to integrate processing and memory, leading to a new field of electronics called “in-memory computing.” A “deep neural network” can have millions of nodes organized in layers to perform specific computations from input data. Integrating logic and memory is a big help.
Bhaswar Chakrabarti, an assistant professor in the Department of Electrical Engineering at the Indian Institute of Technology Madras, is among scientists investigating ways to integrate processing and memory. “I’ve always been fascinated by ‘memory,'” Chakrabarti said. quantum, “in both humans and machines”. So he set about designing a memory his chip that could offer an “alternative computing paradigm” with higher performance and energy efficiency.
Among the various in-memory computing hardware, one that Chakrabarti found particularly interesting was “content addressable memory” (CAM). Basically, when you search a storage device to retrieve information, you don’t search by “address”, you search by “content”. CAM is therefore “a good candidate for a wide range of applications in data-intensive, high-performance search operations,” he says.
Chakrabarti started designing CAMs useful for applications such as network routing, CPU caching and deep learning. His idea was to use a special type of transistor that is of interest in today’s electronics industry: the “ferroelectric field effect transistor” or his FeFET. (A transistor is a part of an electronic circuit that amplifies and regulates the flow of electricity in the circuit.) FeFETs are made using a compound called indium gallium zinc oxide, and are “for deployment in memory.” It is being vigorously researched. Computing”. Chakrabarti procured his FeFET from the Fraunhofer Institute in Germany, which collaborated on this research.
Chakrabarti designed a new CAM cell using FeFET transistors. This “significantly improves density and energy efficiency compared to conventional ‘complementary metal oxide semiconductor’ based cells.” For example, this design uses only one eighth of his transistor in a conventional transistor.
Chakrabarti and his fellow scientists publish paper on their research Applied electronic materials“Simulations show that the proposed CAM has sufficient decision range to perform search operations. Our proposed CAM is very promising as an energy-efficient in-memory computing platform compared to other solutions, because its simple 1 FeFET 1 transistor Because of the architecture and multi-bit operation,” the paper states.
Chakrabarti said more work is needed to roll out chips based on this design in the industry. To synchronize with this new type of chip, memory his array must be developed and peripherals tweaked. Nevertheless, his new CAM is a breakthrough in the field of electronics in the age of artificial intelligence.
This is the last free article.
