Are there any commercial tools for the RISC-V architecture (i.e. where a licence includes full support), or is the Eclipse/GCC toolchain the only option at present?
What kind of support do you want? gcc and eclipse are mature technology and unlikely to be very broken compared to hacked-up proprietary tools, so the main support needed seems to be beginner hand-holding. I don't use eclipse, but gcc, gdb, openocd work fine for me.
We currently have a number of seats of the Keil tools, and I have to say that support has been excellent on the occasions we have needed it e.g. installation issues, target support and even a couple of compiler bugs (a few years back though).
I have nothing against GCC given it's price, and have used it for many years on personal projects. There isn't much love for Eclipse amongst the team I work with however.
The driver here is that timescales are quite tight, we need to be able to hit the ground running on the first iteration of silicon so we don't want to waste any of it trying to resolve any toolchain problems that may arise by ourselves.
Interesting. Who is the ASIC vendor?
Making custom ASICs with a RISC-V core in the Cortex M4 class is exactly the thing SiFive has already demonstrated they can do -- and want to do. Contact them if you haven't already.
I'd rather not say right now, but the cost of ARM licensing has been brought up by most of the smaller vendors we have contacted and RISC-V was suggested by two of them.
That's so cool, what sort of prices does ARM want for cortex M4?
Not as cool as it sounds IMO; it's a huge amount of work with tight deadlines, and a considerable investment for a fairly low volume demand so the risk is high. However we need large reductions in the size of our current discrete solution, and we are already down to 01005 passives and chipscale ICs wherever possible, so not much choice really.
We've not had an official quote yet, but my research suggests around $200k for licensing, no idea on royalties. It's relatively low compared to the total cost of developing the entire thing, but needs to be traded off against possible risks and time of using an unknown (to us) architecture and unsupported tools.