By Dan Jones
Sun Microsystems Inc is targeting the emerging market for media- rich internet and communications applications with a new multi- processor on a single core chip architecture, dubbed Microprocessor Architecture for Java Computing (MAJC or ‘magic’), which is intended to spawn a family of system-on-a-chip (SOC) devices. The project’s lead architect, Marc Tremblay, a distinguished engineer at Sun, says that, initially, the design will be aimed at multimedia devices using MPEG streaming, voice recognition and 3D rendering and also in wire-speed communications infrastructure hardware.
The concept of the MAJC design is simple – multiple high performance processor cores on a single die – which can be programmed in a Java environment. Sun’s Bill Joy first hinted at the design at JavaOne in June (CI No 3,684). The device is built around UltraJava cores that execute 128-bit-wide very-long- instruction-word (VLIW) instructions. The design differs from other system on a chip designs – like National Semiconductor Inc’s Geode design, because instead of using a central CPU teamed with other application specific blocks – for instance, a graphics accelerator, the MAJC concept uses general purpose, high performance processors performing multiple tasks. Tremblay explained that the inherently multithreaded system could feature one processor running the operating system and another running an application or have the have the processors joined at the hip collaborating on applications. This design takes advantage of the multithreaded nature of the Java language itself. Tremblay expects that the design will double the speed of applications running on it. Jeff O’Neal, group marketing manager for MAJC at Sun, said that the use of Java was crucial to the project. With Intel Corp’s IA-64 architecture just around the corner, he said, the last thing that ISVs want to deal with is another instruction set. In addition, the most recent version of Java has incorporated a lot of the multimedia support – 2D and 3D graphics, speech recognition – that developers would need to use designing applications for internet appliance boxes.
Sun says that it will initially look for the chip to be used in advanced set-tops. MAJC chips, Tremblay said, would allow for creation of a service portal in your home. However, neither he nor O’Neal would be drawn on exactly what this meant. The likely answer is that Sun, through its link with America Online Inc, will use the chips to create extremely powerful set-top boxes for the ‘AOL TV’ project. In-car computing devices would be another pervasive application, according to O’Neal. However, Sun is not just looking at the consumer market for the new design but is seeking to broaden its reach in the data networking and telecommunications market. Tremblay said that in the design, all the data types were held in the same register and multiple data types could be processed. These kind of features would lead to wire-speed communications devices – hardware that pushes out data at the top speed of the network without impediment from the processor. This ability combined with video streaming could finally make video conferencing the kind of viable application that it has threatened to be for years.
However, all of this is not long off the drawing board. Tremblay will present a paper on the design at the Hot Chips conference later this month and disclose technical details at the Microprocessor Form on October 4. The first chips, which are likely to arrive next year will feature two processors on one piece of silicon and run at 500MHz to 600MHz. In theory, the design is scalable to thousands of processors on a single die, however, Tom Halfhill, embedded analyst at the Microprocessor Report, expects that that point is ten or more years away. Tremblay expects processor density to increase four-fold every three years, in accordance with Moore’s law.
Sun is wary of describing its new design as a multimedia chip. However, it does share some features with ill-fated designs such as Chromatic Research’s Mpact chip. Chief among these is the use of a VLIW instruction set. VLIW instruction sets have a reputation of being tough to program. However, Halfhill says that VLIW has a bad image because it is hard to write compilers for it. Tremblay agrees and says that his design team has been working hand-in-glove with the compiler team during the four-year design period. He also says that the Sun design is much simpler than previous VLIW efforts.
Another problem for Sun is its heritage as a workstation and server vendor. Other embedded chip designers have said that Sun only developed chips suitable for the very high end of the embedded market, which although powerful, consumed a lot of energy. Tremblay, who is a veteran of the UltraSparc development team, says that Sun was very aware of the power issue and that the scalability of the new design meant that chips could be developed to span a broad range of embedded applications – from battery powered handhelds to high-end communications hardware. In theory, the chips could also serve as a rival to Intel’s IA-64 chips. However, O’Neal said that Sun wasn’t looking to move into that market with the new design because Sun already offered the Sparc processor line for servers. He also suggested that server vendors and ISVs wanted continuity and wouldn’t be willing to move to the clean sheet of the MAJC architecture. Halfhill was less diplomatic, suggesting that Sun would be foolish to get in the path of the Intel juggernaut in this established marketplace.
