Site: Mitsubishi Electric Corporation
Semiconductor Research Laboratory
Amagasaki, Hyogo 661 Japan
Date Visited: November 16, 1994
Report Author: R. Hickernell
The Semiconductor Research Laboratory, with 120 researchers, is one of the 10 corporate research and development laboratories of Mitsubishi Electric Corp., and one of four in the Amagasaki region of western Japan, near Osaka. The Central Research Lab and the Materials and Electronic Devices Lab are located on the same site. The Electro-Optics and Microwave Systems Lab is located in the Kamakura area. Corporate R&D laboratory researchers number about 2,500 in total. Optoelectronics development is also pursued in the Optoelectronic and Microwave Devices Lab of the Semiconductor Group. The Semiconductor Group has 1,100 researchers located in the Kita-Utami area. Mitsubishi Electric Corporation has 107,800 total employees.
The Semiconductor Research Lab was formed as a result of laboratory's restructuring in June 1993. It has three main purposes: semiconductor (mostly silicon) processing and materials; next-generation Si and III-V semiconductor devices; and new device designs for VLSI, parallel processing, DRAMs, and OEICs. The Advanced LSI System Technology Department, with 30 researchers, is charged with the third purpose. Researchers from the department gave presentations on the development of an artificial retina and an optical neural chip.
Optical neurochips typically require LED, photodetector, and spatial light modulator (SLM) arrays to accomplish image processing and display functions. The variable sensitivity photodetector (VSPD) concept developed at Mitsubishi essentially combines the SLM and detector functions in a single device. The GaAs version of the VSPD uses interdigitated electrodes on a GaAs substrate to vary the detector sensitivity by varying the applied voltage.
The artificial retina consists of a two-dimensional array of VSPDs whose adjacent elements are set to predefined sensitivities (depending on the intended function) before scanning an image. Functions already demonstrated include edge extraction, smoothing, correlation, and pattern recognition. Dr. Kazuo Kyuma et al. hope to develop an "intelligent camera" product based on the artificial retina. Targeted applications are in the automotive, aircraft, military surveillance, and optical character recognition industries. A prototype has demonstrated 80 x 80 microns resolution and a frame rate of 2.5 kHz. Two stated advantages of the device over other groups' versions of an artificial retina are the use of GaAs photodetectors and the ability to perform several different functions using the same array.
The artificial retina has been applied to Japanese kanji character recognition by a visiting researcher from the United States working at Mitsubishi. Using a clustering algorithm, the number of pattern comparisons for the standard Joyo kanji set can be reduced from 1945 to 147 and still retain a 99.99% success rate of recognition at a speed of 2.3 classifications/sec.
Dr. Anno Hermanns, also a visiting researcher in the department (former postdoctoral researcher at the University of Colorado), described his work on a silicon bipolar VSPD. The challenge has been to retain the advantages of low cost and high density, which is difficult for CCD or MOS imagers with on-chip processing. The solution is a lateral p-n-p structure fabricated using a standard 1.2 micron CMOS process. The 23 x 23 micron pixels have high photosensitivity (30 A/W), fast response time, seven orders of magnitude dynamic range, and are easily integrated with other analog CMOS circuitry. Minimum detectable light levels are approximately 1 uW/cm 2. A 128 x 128 prototype array of Si VSPDs is currently being fabricated by a company in California. The 18 x 18 device array reportedly had no problems with crosstalk.
Dr. Kyuma stated his personal objectives for the artificial retina chip: first, to see its application to a product within the company, and second, to offer it through the United States-Japan Joint Optoelectronics Project. He mentioned two methods of moving products from R&D to commercialization: staff from the R&D group move with the product to the business unit, which works best if the market is large, or the business unit goes to the R&D group and learns the technology, which works best if the market is small. Dr. Kyuma personally believes that he must now have a significant role in the marketing of his division's products, a break from the past when a marketing group found most applications for the R&D groups.
Dr. Kyuma's department is receiving about 50 million yen from the Real World Computing project, primarily for labor, for work on computing applications. He believes that the optimum funding balance for a project in R&D is 50% funding from corporate R&D and 50% from the operating units. At the present, his corporate R&D budget share is larger than 50%.
According to Dr. Kyuma, Mitsubishi Electric as a whole wants to be strong in systems applications, as demonstrated by the names of the company's laboratories. Due to low profit margins for individual devices, the only devices that tend to make profit in the electronics business are DRAMs and microprocessors. He feels that, in Japan, the number of companies that have very similar sets of products is too large, causing intense competition and a smaller profit margin. Differentiation in product lines and systems applications of electronic devices are two paths to profitability. Mitsubishi foresees selling VSPD arrays as a key part of a vision system, not as individual chips.
Dr. Hermanns stated that there are no silicon foundry services in Japan as there are in the United States. Japanese university researchers are particularly disadvantaged by this situation because of the low funding levels for in-house facilities and the difficulty of accessing industrial facilities. The Japanese government is trying to start a domestic foundry service.
The JTEC panel's visit to the Advanced LSI System Technology Department focused on longer-range research and development of optoelectronics for image processing and computing. The work is of high quality and has some practical goals for application. The panel was not able to visit the nearby Optoelectronics Laboratory, which is responsible for the development of III-IV compound semiconductor devices such as GaAs-IC and laser diodes.