| ??? 12/11/08 17:04 Read: times |
#160864 - Would that be appropriate on 805x? Responding to: ???'s previous message |
Per Westermark said:
A 10k run can be a small run, if the development costs of the firmware is $1,000,000+ in which case the sw cost is > $100 / unit. A 10% change in development time would then represent $10/unit which may possibly be twice the price of the microcontroller... If the software cost is so large, it's unlikely one would use an 805x, though, isn't it? Wouldn't you be more comfortable using, just as one example, an ARM9 CPU, one with a built-in SDRAM channel and built-in FLASH/SRAM channel, along with the "usual I/O" and supported by GNU development tools and native LINUX? That would probably reduce development cost by an order of magnitude. From where I sit, it's hard to imagine a product that would have a $100/per item firmware cost at 10K piece volume, but that doesn't mean they don't exist. With a "large" CPU, equipped with $2/16MB SDRAM, you could do lots of things you wouldn't attempt with 805x using compilers designed for small MCU's. Not all embedded equipment have trivial software. Quite a lot of products sets the customer price based on the know-how in the software, not based on the hardware cost.
And such know-how-based products are more likely to move between different hardware platforms which greatly affects the reason to have code in asm or in C. I'm a bit scared when I see people putting forth too general claims. Arguing too general claims in one situation probably means arguing too general claims in other situations too. What I like about developing software is that most products are different. I don't have to implement the same wheel all the time, but have to figure out new "recommended" solutions for each and every product. In the end, my designs will have to contain more or less guesses depending on how much facts are known (even by the customer) and my own knowledge in the relevant area. But since I'm involved in the hardware and coding, the quality of the design will directly affect myself forming a feedback loop that makes sure I really do my best with the design. What would cause a firmware development effort to cost a million bucks? I'd guess, just over the thumbnail, it might be $10K for algorithm development and testing, and the rest for making it fit in the small 805x environment. After all, firmware development cost is tools and man-hours. If the algorithms are easy to develop and test under LINUX, then having used an ARM or MIPS has the advantage that, once proven, these algorithms are directly portable to the target, leaving only the I/O to be worked out on the target environment. RE |
