Backers of Open-Source Chips Launch Startup
wsj.com
wsj.com
What astounded me was that Intel post 2008 and AMD post 2011 have seperate management chips that mean essentially any motherboard is rooted at the start.
I still need to dive into the complexities (I knew about Intel ME but assumed we could get round it somehow) but we need these open chips more than ever.
http://spectrum.ieee.org/semiconductors/design/the-death-of-...
Industry has a vested interest in maximizing returns. This means that it tends to pump out very practical, usable products, but often gets stuck at local maxima. It makes more sense for Intel to keep hammering away at x64 than to switch to a possibly better architecture, even if switching might be better in the very long run. OSS is driven not only by industry, but also by academics (who have a vested interest in doing highly experimental research) and hobbyists (who have a vested interest in quality and simplicity). This additional set of interests means that OSS is being pulled in more directions and is more likely to escape local maxima. If software was entirely driven by industry, we'd probably still be using FORTRAN or something.
Now, the problem is that open source projects tend to have fewer resources and less time pressure, so development is slower. This is fine for computer science, where an academic who develops a novel language or algorithm can reasonably expect it to be cutting-edge for quite a while and still in use decades in the future.
On the other hand, processor design moves so fast that the only way to keep up is to dump tons of money and man-hours into pursuing the latest and greatest as aggressively as possible.
As the pace of processor development slows down (if it does), it will give a chance for the slower, but higher-quality, work of academics and hobbyists to catch up.
I am looking forward to all sorts of esoteric architectures implemented on all sorts of esoteric theories to come out of the woodwork. We're very likely to find some good stuff that we haven't thought of yet.
You're right that the hardware field is totally different from the software field, however you have to see that instruction sets, which is what RISC-V is about hardly ever change. Today's x86 processors are still compatible with 8086 from 1976 (and assembly compatible with 8008 from 1972). Similarly ARM is from 1985.
The thing that's keeping hobbyists out is that producing ASICs is prohibitly expensive.
We may see some improvement in performance due to the new architecture, and maybe even more so with the Mill CPU architecture. But the end of the road for silicon lithography already in sight, and these new architectures won't fix that.
The reason is simple, even a single manufacturing fault can make a chip unusable. If you assume e.g. a fixed number of faults per wafer, you can easily see how increasing the chip area increases the area you have to throw away for every fault.
As you wrote, the semiconductor business is very capital intensive, so no one can afford to do that (except for some niche applications, like R&D, space, etc.).
But it's still an effort to share knowledge, which is good.