1. 14
  1. 3

    The patent bit is particularly interesting. When we started CHERI, we used MIPS III as the base ISA because it was 20 years old and so there existed at least one possible implementation that didn’t trample anyone’s patents. I have no idea if we did infringe any patents (our branch predictor was more clever than 1992 MIPS cores and our TLB was optimised for FPGA, so it’s possible that we did) but no one cared because we were a research project producing a prototype softcore.

    Intel and AMD have a massive patent portfolio. Arm maintains one for the partnership that every partner has access to for Arm cores. The odds of building a high-performance implementation of a CPU without accidentally infringing at least one patent is incredibly low. While RISC-V is fairly niche and while it is mostly deployed in simple core designs, it’s safe, but as soon as it starts to cut into business for the big players then expect the lawsuits to start.

    The nice thing about hardware, from the perspective of a patent troll, is that anyone who can bring a competitive chip to market needs a budget of at least $20m, so they’re high-value targets compared to most open source software projects.

    1. 1

      Anything used in Pentium Pro to Pentium III is out of patent protection. Pentium M and Opteron are only a bit over a year away from their 20th anniversary of commercial introduction.

      I don’t know, but I’d imagine SiFive might have some kind of patent licensing deals with their investors Intel, Samsung, Qualcomm.

      MIPS have a pretty extensive patent portfolio and will be making high end RISC-V cores.

      1. 1

        Anything used in Pentium Pro to Pentium III is out of patent protection. Pentium M and Opteron are only a bit over a year away from their 20th anniversary of commercial introduction.

        True, though note that this does not mean that any implementation of the same ISAs will be. There are a lot of patents covering SIMD, virtualisation technologies, modern branch predictors, register rename, prefetchers, and so on are covered by a huge number of patents and a lot of those are in the last 10 years.

        Of course, it’s always possible to build cores based on 20-year-old designs and maybe that’s fine in a lot of places.

        I don’t know, but I’d imagine SiFive might have some kind of patent licensing deals with their investors Intel, Samsung, Qualcomm.

        Which would be fine for SiFive, not so fine for anyone else in the ecosystem. Even that doesn’t help too much because a lot of the patents that Samsung and Qualcomm own build on patents from other members of the Arm partnership and so even if you have a license for the Samsung / Qualcomm patent you might not have the rights to actually implement it (patents stack, so you can patent X and separately patent and improvement on X and having a license to the second patent doesn’t let you implement X + the improvement unless you also have a license to the first).

        MIPS have a pretty extensive patent portfolio and will be making high end RISC-V cores.

        Actually, they don’t. When MIPS was sold to ImagTec, a consortium of silicon vendors bought the patent portfolio because they didn’t want it to end up in a patent . This is the only time that Arm has ever bought patents from a third party (and then, only as part of a consortium). ImagTec received a license to these patents but cannot use them defensively because they don’t own them and so cannot enforce them (the companies most likely to sue new silicon vendors jointly own them). Subsequent owners inherited only what ImagTec owned. Most of the MIPS patents will be expiring soon, so this doesn’t give MIPS a particularly useful advantage.