07.25.18
Applied Materials, Inc. announced it has been awarded a contract by the Defense Advanced Research Projects Agency (DARPA) to develop a new type of electronic switch for artificial intelligence (AI) that mimics the way the human brain works to enable dramatic improvements in performance and power efficiency.
The project is being supported by DARPA’s Electronics Resurgence Initiative, a multi-year research effort intended to achieve far-reaching improvements in electronics performance well beyond the limits of traditional Moore’s Law scaling.
Applied is working with Arm and Symetrix to develop a new neuromorphic switch based on CeRAM memory that can allow data to be stored and processed in the same material. The goal of the project is to enable a major improvement in artificial intelligence compute performance and power efficiency with the use of analog signal processing as compared to current digital approaches.
“This project is a perfect example of how new materials and architectures can be developed to enable new ways to accelerate artificial intelligence applications as classic Moore’s Law scaling slows,” said Steve Ghanayem, SVP of new markets and alliances at Applied Materials.
The Applied Materials team is part of the ERI Foundations Required for Novel Compute (FRANC) program, which seeks innovations that go beyond von Neumann compute architectures. Central is the design of circuits that leverage the properties of new materials and integration schemes to process data in ways that eliminate or minimize data movement. The novel compute topologies that come out of this effort could allow processing to happen where the data is stored with structures that are radically different from conventional digital logic processors, ultimately allowing for significant gains in compute performance.
The project is being supported by DARPA’s Electronics Resurgence Initiative, a multi-year research effort intended to achieve far-reaching improvements in electronics performance well beyond the limits of traditional Moore’s Law scaling.
Applied is working with Arm and Symetrix to develop a new neuromorphic switch based on CeRAM memory that can allow data to be stored and processed in the same material. The goal of the project is to enable a major improvement in artificial intelligence compute performance and power efficiency with the use of analog signal processing as compared to current digital approaches.
“This project is a perfect example of how new materials and architectures can be developed to enable new ways to accelerate artificial intelligence applications as classic Moore’s Law scaling slows,” said Steve Ghanayem, SVP of new markets and alliances at Applied Materials.
The Applied Materials team is part of the ERI Foundations Required for Novel Compute (FRANC) program, which seeks innovations that go beyond von Neumann compute architectures. Central is the design of circuits that leverage the properties of new materials and integration schemes to process data in ways that eliminate or minimize data movement. The novel compute topologies that come out of this effort could allow processing to happen where the data is stored with structures that are radically different from conventional digital logic processors, ultimately allowing for significant gains in compute performance.
