Slashdot Mirror
Five PC Vendors Face Patent Lawsuit
Combuchan writes "This article from internetnews.com caught my attention: While Linux lawsuits gobble up the IT community's mindshare, a lesser-known legal action is being fought seeking billions of dollars from five PC vendors. Patriot Scientific, a small, San Diego-based seller of embedded microprocessors for automotive and scientific applications, is suing Sony, Fujitu, Matsushita, Toshiba, and NEC, alleging infringement of a Patriot patent for what it calls 'fundamental microprocessor technology.'"
Had anyone else not heard about this company before today?
Patent No. 6,598,148,
appears to be a divisional patent from patent No. 5,982,231, filed in 1989. In summary, the invention describes a CPU using both stack and register-based instructions which interfaces directly to DRAM memory over 32 shared address and data lines. Memory accesses are minimized by using mainly stack-based instructions and accessing memory in fast page mode. The invention is designed to minimize cost, fitting just tens of thousands of transistors into a 44-pin package, while delivering approximately 20 MIPS performance. While none of the design elements by themselves are new or unique, the particular combination chosen seems to fit the definition of a real invention that was not in existence before 1989.
The features which seem to relate to its claims of variable speed are the use of an on-board ring oscillator to drive the system clock, and memory cycle timing which automatically adjusts to the loading seen from attached memory devices. Using an on-board ring oscillator means that the system clock can be driven by a device built with process parameters identical to other devices on the chip. If CPU transistors are a little slow on a batch of chips, the ring oscillator transistors will be similarly slow and the chips will continue to work at a slower but reliable speed.
It's just not clear to me how Patriot Scientific can claim that companies like Intel are infringing on their patent. Is it because Intel's products contain technology which also happens to appear in their invention, but which they themselves did not invent? None of the individual technologies used in the Patriot device originated with them, and it appears that only the specific combination of those technologies described in the patent application would have qualified it for approval. However, the patent claims are written in such a way as to seemingly describe a much broader range of CPU designs, many examples of which were certainly in existence before even the original patent application by Patriot was filed.
I've seen it stated that the claims are the only important part of a patent. That seems to be the big problem with this particular patent. What happens when a legitimate invention is granted a patent with overly broad claims? Some patents are written with claims that include a range of claims, some very specific and some very broad, on the assumption that it's better to aim for as much as possible and settle for what you can get. This particular patent seems to have only broad claims. Does that make it easier or harder to try and overturn?
The patent application is very specific about using a ring oscillator to control the processor clock. Most processors use PLL (phase locked loop) devices to control their clock frequency.
Typically only asynchronous CPUs are clocked by free-running ring oscillators. Almost all CPUs on the market, including embedded CPUs, are synchronous designs and they come in fixed speed grades, which makes PLL clocking very attractive.
Even with PLLs you can vary the clock speed, e.g. when you detect that the chip is too hot or when the work load is low the CPU clock can be scaled down in discrete steps.
A ring oscillator provides a non-discrete (continue) frequency range, but on the downside its clock frequency is very unstable compared to PLLs, which is bad for synchronous designs.
There is also prior art: the early MIPS processor implementations had a free running multiplier, which operated as fast as the silicon would allow, even though the rest of the CPU was clocked at a fixed frequency.
In summary: this patent is not worth a lot.