Issue No.06 - December (1995 vol.15)
Published by the IEEE Computer Society
Since microelectronics development in Japan takes place in many different fields, I limit this discussion to LSI chips for computers in microprocessors, memory, and peripheral chips. VLSI microprocessors and memory are especially important, incorporating technologies that are basic to microelectronics.
No longer, however, is VLSI microprocessor development simply a matter of coming up with the best technology. Success comes to whichever company can come out the quickest with the most capacity for the least money. By comparison, fields like memories and disk drives still show all the earmarks of a technology battle among developers--perhaps because development is still driven more by what is possible than by markets. In the more need-driven VLSI microprocessor market, customers who adopt a processor for applications determine the demand. For example, the determining factor for peripheral chips is customer acceptance of the specifications adopted in the chips. (This applies to both original and standard specifications.) Besides technology, economic and political factors strongly influence choices here.
When we think of microprocessors, the first big market that comes to mind is that of personal computers and workstations. This is undeniably a major source of demand, with annual sales in the tens of millions and a growing trend to replace large-scale computers. The question of which microprocessor gets adopted clearly has to do with market and political forces and not only technological superiority. In fact, the fastest processors are not the ones that sell the most.
In this field Japanese makers have lost out entirely, though they have the capability to design and develop new architectures. As of 1993, Hitachi's Gmicro/500 processor developed on the TRON architecture could boast of being faster and consuming less power than the Intel Pentium, which appeared around the same time. 1 Developing an original architecture from scratch is one thing, however; gaining acceptance for it is another. What it comes down to is that without a large market share there is no way to compete on cost. And right now Japanese makers have no plans to build their own personal computers or workstations from zero. In this field, research and development on a processor with a new architecture remains in the labs, with almost no development aimed at practical implementation. The only exception is work on super high-speed chips for use in supercomputers.
One more big market for microprocessors is embedded systems. Because Japan has essentially no military market, 2 development focuses on consumer and industrial equipment. This explains why microprocessors for embedded systems and consumer markets have long been a target of active development. The end products are not computers. For this reason manufacturers that aim microprocessors at the embedded market do not have to worry about compatibility with legacy software as in the computer market; nor is it necessary to provide a wide array of application software to attract customers. It is thus an easier market to penetrate than that for personal computers and workstations. On the other hand the cost requirements are severe, a few dollars per chip or slightly more, making large-lot sales a prerequisite.
In spite of the difficulties, microprocessor development for embedded systems proceeds at a furious pace. At the high-performance end of 32-bit microprocessors is NEC's V800 series, Hitachi's SH series, Mitsubishi's M16 and M32 series, and Fujitsu's FR20, among others. Performance ranges from a few dozen to 100 MIPS or more. Applications targeted include 32-bit game machines, car navigation systems, and PDAs (personal digital assistants).
A fairly substantial market is foreseen for each of these fields. Sega's 32-bit Saturn game machine, for example, runs on two of Hitachi's SH2 chips. Sales of 32-bit game machines for home use can amount to several million units a year, which is a rather large market even for an embedded MPU. Although these high-performance embedded microprocessors are touted as RISC chips, their being designed for embedded use means they typically adopt measures such as use of variable-length instructions or 16-bit instructions for the sake of efficient object code. In many cases they incorporate on-chip multiplier/accumulator units or DSP functions aimed at multimedia applications. And most of these chips support ITRON or μITRON 3 operating system specifications. This architecture is widely adopted in Japan for real-time operating systems in embedded systems (see Figure 1). Besides being an specification, the architecture features the compact, lightweight design needed in embedded systems.
This issue introduces the latest microprocessors by NEC and Hitachi. Both companies' products are RISC chips for embedded use, featuring high performance relative to their low power consumption. NEC's V830 is a low-cost chip with built-in DSP functions, aimed at multimedia applications. Hitachi's SH family of processors is adopted in digital still cameras and several other consumer products in Japan besides the Sega Saturn. The Hitachi article focuses on the SH3 series, which is geared to PDAs.
Japan is not very active in developing peripheral logic chips for personal computer use; Taiwan now ranks alongside the United States as a major development center. Japanese firms are actively developing chips as standards are decided: JPEG and MPEG for multimedia use, or ATM and the like for telecommunications. Part of the reason is that these Japanese semiconductor companies manufacture electronics products across a broad range, including whole systems. Demand in this area will likely increase steadily; at the same time there is plenty of room for venturing into the internal architecture to realize performance enhancements, so long as the standards are kept to, giving this field lots of appeal. This issue introduces Fujitsu's 3D graphical LSI chip set with performance of 300,000 polygons per second. A fascinating aspect of this set is that adding more chips further increases drawing performance.
The technological goals here remain, above all, greater capacity and speed. From a cost standpoint DRAMs continue to gain in overall share. While Japan still leads in production, South Korea is rapidly closing the gap. In producing 16-Mbit DRAMs, Korean makers have taken over the lead. Along with greater capacity, users are demanding high-speed DRAMs with which they can interface without the use of costly SRAM, to match the speed of today's CPUs. In this connection I've included a survey of recent high-speed DRAM interfaces in general, without limiting the discussion to Japanese technology.
Looking back in time, at least in terms of computers, one wonders if there was ever an age when political and economic factors held such sway as at present. In the age of the mainframe, from the 1960s through the early 1980s, a different manufacturer meant a different architecture. There were those who came out with IBM 360/370 compatibles, but they were a minority. Likewise in the minicomputer era represented by DEC's PDP-11 and VAX series, different architectures abounded.
After IBM brought out the PC, however, things changed, so that today the architecture battles are all but over. For workstations, only one or two architectures are likely to survive. All of these architectures are American. Like a microcosm of the political world, the only global superpower able to exercise its sovereignty is America. Japan can never become the leader, nor does it have the will to be.
As can be seen in DRAM production, Japan is a master at turning out products of uniform quality at low cost, even with high labor costs. There is a limit to such techniques, however. Countries like South Korea, Taiwan, and Singapore are catching up fast, and may even surpass Japan some day.
Aware of this reality, Japan is now particularly interested in developing new application fields in as yet unexplored territory. This is the approach being taken by the TRON Project, which I have introduced on numerous occasions before. 4-6 In existence for more than a decade, the project aimed from inception to develop technology for building "computing-everywhere" environments, with countless computer chips embedded in living spaces. Ten years ago many people considered such thinking to be crazy, the stuff of science fiction; but today it is seen as a realistic target. In the 1990s the US became the scene of active research efforts turned toward similar goals, under themes like ubiquitous computing and computer-augmented environments. 7
In spite of political and other obstacles over the past decade, TRON approaches its final stages. Our aims are open, highly functionally distributed systems. Millions or billions of computers can be loosely coupled, enabling them to collaborate. We have built automatic configuration, fault tolerance, and other aspects of the technological backbone. Results like multilayered design diversity have also emerged. 8
The results of 10 years of work will further come to fruition in the TRON Hyper Intelligent Building, due to start construction in March 1996. It implements the computing-everywhere environment on a real-life scale. An article in this issue discusses real-time operating systems for this kind of environment.
Computer architecture may appear stagnant at this point in time. In fact, however, new applications continue to emerge and forge new architectures, because existing architectures are not up to the job. As the micro world goes through dramatic changes, it will be essential to create new applications and the new architectures for them. Instead of simply getting excited about Windows 95, now more than ever we need applications offering a new set of challenges. We believe for this reason TRON is in a position to contribute like at no other time.
I would be very pleased if this special issue helps you understand the approach being taken by the microelectronics industry in Japan.