S?bastien Bourdeauducq wrote: > Let's fix that problem instead. Developing a reference design for a > proprietary chip is an experience you can have at any semiconductor > company.
I think mid-range SoCs are becoming a commodity. As developers who cherish their independence, the best we can do is to travel lightly, so that we can pick the chips we can readily use and don't lose time daydreaming about things we can't have. This means, among other things, not to lock ourselves into proprietary features. It also means to try to do as much of the work we need done at layers there the subtle but in the end irrelevant differences between chips don't bother us. One may view picking an off-the-shelf SoC as a step back from the freedom we've obtained with Milkymist, but I think it would still be a very reasonable move at the present time. One issue with an FPGA-centric design, like in Milkymist, is the extreme slowness of the CPU core. An extremely slow core means that a lot of "standard" software will have trouble running satisfyingly. An FPGA-centric design can still run circles around anything else when it comes to certain specific tasks implemented in "gateware", but for a Nanonote intended as a general-purpose device, it would be a problematic choice. You once suggested one could add a custom ASIC just containing the CPU core, with the FPGA taking care of peripherals and the memory interface. I find this a very interesting idea. But we also have to recognize that we're currenly not in a position to actually make such a project happen. But we can save this idea for later. The work on Milkymist, past and ongoing (your memory controller, Yann's MMU, Lars' renewed Linux effort) is not lost. Instead, it closes gaps we have already suffered from in the past. When the day comes to try something FPGAs-centric again, perhaps even with a helper ASIC, having already done these things will greatly accelerate the development process. It's much easier to climb Mt. Everest when you start from a camp at 7000 m than from sea level. Meanwhile, we still haven't fully tapped the potential of the Nanonote concept. Doing that calls for a more evolutionary than a revolutionary approach. If a chip maker were to have a sufficiently common interest with us in making this happen that they could sponsor such an effort, I wouldn't rebuff them too hastily. - Werner _______________________________________________ Qi Hardware Discussion List Mail to list (members only): [email protected] Subscribe or Unsubscribe: http://lists.en.qi-hardware.com/mailman/listinfo/discussion
