Ceva claims to have a dozen design wins for its Ceva-XC DSP. As it stands today, they include ZTE and Mindspeed, and one other "infulential wireless infrastructure supplier," which Ceva's CEO talked about during the recent analysts call.
Frinancial analyst Gary Mobley at Benchmark recently speculated that "influential wireless infrastructure supplier" is Ericsson. But CEVA is not confirming or denyaing it for now.
You have a point there. But Ceva might disagree with your statement about "one large Chinese telecom OEM doing ASIC doesn't mean much."
Ceva claims the number of system OEMs wanting to license cores ARE increasing in recent months. That, in their eyes, is a "big change" unfolding in the industry at the moment.
Well, one example of ASIC being design by a large Chinesse telco doesn't mean much, they just probably want to lower the unit cost...large ASICs are extermely complex to design properly...it takes time (1-2 YEARS, sometimes longer) and money (at least $5M, sometimes $50M)...you need a billion dollar market to justofy the cost...sure, there will be ASICs always designed for these large socket opps but the number of ASIC tape-outs has, is and will be diminishing...sad reality for ASIC design guys (myself included ;-)
It's not a slam-dunk for the tier one guys either. I used to work for one of the major defense contractors, and one of their projects was at one point 2 years behind schedule because they couldn't get an ASIC design to work. The problem essentially was that they had a bad system design (VME really doesn't work very well when you extend it 50 feet or so... Who'd have thunk it?). Just because you are a big company doesn't mean that you do all things well.
so you are saying the death of ASIC hsa been greatly exaggerated. I guess I am one of those as guilty as charged.
That said, though, I'd have to wonder that this trend for home-grown SoCs must be limited to tier-one OEMs. Unless you have a clear ROI in mind, it's still a risky proposition. No? I mean, to do it inhouse.
@Larry, that's fascinating...so the tools available today are definitely changing the landscape, you say.
But as you also pointed out, it still takes a team of talented designers to develop a home-grown SoC, which is probably no easy feat.
One question. Ceva executive was telling me that if you don't have a in-house SoC design capability, you could still ask an ASIC vendor to build your SoC for you. I wonder which ASIC companies are profiting from such a trend...
It used to be that the sheer complexity of the design process and the toolchain support made this kind of effort a bad idea, but I'm not so sure that is true anymore. We are using an Atmel SoC in one design, but we used buildroot to configure and build a customized OS and cross-compiler in a very short period of time rather than using the toolset that Atmel provides. That way we get to decide exactly how fast we want to integrate new changes. I can see moving from that to licensing the ARM core and custom-tailoring the peripheral interfaces ourselves. Unfortunately, I can also see companies getting in deep trouble trying to do that with engineers that are not up to the task.
Blog Doing Math in FPGAs Tom Burke 20 comments For a recent project, I explored doing "real" (that is, non-integer) math on a Spartan 3 FPGA. FPGAs, by their nature, do integer math. That is, there's no floating-point ...