TOKYO, February 14, 2018 - Mitsubishi Electric Corporation (TOKYO: 6503) announced today it has developed a compact hardware artificial intelligence (AI) technology that can be implemented on small-scale field programmable gate arrays (FPGAs) that users and designers can configure to develop tailored applications. Mitsubishi Electric leveraged its Maisart*-brand AI technology to achieve a solution that uses a highly efficient calculation order and optimized circuit architecture for improved real-time processing and reduced costs, which is expected to expand the scope of AI deployment in home appliances, elevators, high-precision maps and more.

* Mitsubishi Electric's AI creates the State-of-the-ART in Technology

Envisioned applications for Mitsubishi Electric’s Maisart-brand “compact AI” solution

Key Features

1)
Compact AI enables fast real-time processing in FPGAs
- Improved calculation-order efficiency and optimized circuit architecture for FPGA implementation.
- Reduces inference computations to one-tenth** the time of conventional AI and improves real-time processing.

** Compared to Mitsubishi Electric's existing technology that implements conventional deep learning on FPGAs
2)
Compact, small-scale circuit architecture will expand usage of AI in FPGAs
- Miniaturized circuit architecture realizes compact AI for implementation in small-scale FPGAs, as well as embedded CPUs, for reduced costs and energy consumption.
- Small-scale FPGA implementation will broaden scope of AI application in home appliances, elevators, high-precision maps and other products/fields where AI usage has been limited by high costs.

Development Summary

  Processing time on the same scale FPGA FPGA scale required for the same level inference Bands/standards
Proposed Compact AI on FPGA Short(1/10th of conventional time**) 3
Conventional Technology 12 dB 35% 1

Note that the releases are accurate at the time of publication but may be subject to change without notice.