You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
Microcontrollers are such a weird market. On the surface , it seems that there's plenty of competition.
But if you look at the work the software developers for this do, a large part of it is about saving ram/flash/peripherals - which are basically almost free to manufacture (and many aren't that hard to design - heck why should an mcu with 4 UARTS cost more than one with 2 UARTS assuming same package? ) but relatively expensive to buy - in volume.
This is the sign of not a very competitive market, to my eyes.
And it was even worse , before ARM , but ARM is a proof that it can be improved and such improvements are a great business opportunity.
So how can it be improved ? both by new products/tech and mostly by new business models ?
ippisi - might this have something to do with regulation and reliability? I remember a few years ago the automotive controller guys were very reluctant to do a lot of work at the design level because that might trigger need for re-certification. I don't know how widely that applies in the general uC market.
I think that this MCU market looks weird... because we can't really talk about one MCU market. It's more like the aggregation of multiple markets, the only common point being: MCU device = CPU + peripherals
This market starts with ultra-low cost segment (PIC or AVR from Atmel) with ASP = fraction of $1, going up to $5 MCU...
Instead of one MCU market, we should say multiple application specific segments. In fact, MCU chip makers tend to design ASSP (Application Specific Standard Part) rather than general purpose products. This seems to be the winning strategy: propose one product for one application, integrating application specific peripherals and the ARM core delivering the MIPS count needed to support this application, integrated in the package (ball pitch, size, thickness, price) which best fit the customer needs.
Eric - I agree , the strategy in the current market works , and also kind of keeps a competitive balance between companies. and don't forget the common use of binning - which also helps.
But let's say we design somewhat generic 65nm mcu die on 300mm wafers, offering 64KB flash and 16K ram and rich variety of peripherals + an extra low-end core or two to act as programmable peripherals.Make it decently low power.
In short not the best mcu , but good enough for a lot of tasks.
The die cost for something like this would be low, maybe 4 cents, right ? and say yield issues make it 5 cents.
So now let's have a simple business model - instead of binning etc , just package this in the different required packages and charge so our gross margin is 50%. So our die only contributes 10 cents to the final cost.
Now , would it be attractive ? well first, this reduces logistics and software development costs for companies, and offers TTM.. For some segments like the industrial segment or the aggregate small vendor segment , saving an extra 5 cents in the bom might not matter. Heck it might be even cheaper for customers than the ideally optimized chip.
Don't you think such a strategy can have a decent success ? And do you think this is the optimal strategy if you're interested in profits ?
I agree with Eric, there are segments. There is a 25 cent segment - and it won't go any lower because nobody can make it up in volume - and a $1 segment and a $5 segment. Most of the MCU vendors are trying to add features to get up into the $5 segment again.
An interesting article crossed my desk this morning about the "new" automotive chip companies - NVIDIA, Samsung, and LG. Those guys have no interest in $1 and down.
Another interesting development was MediaTek announcing a new IoT chip last year with an ARM7TDMI core because it was sunk licensing cost and dirt cheap where not a lot of performance or power management was needed. Back in the day, ARM7TDMI was an MCU core before Cortex-M.
Chasing these explosive growth opportunities where volumes rise and costs fall rapidly is fantasy. Sustained CAGR over 20% is not realistic. Use mature nodes to make money and stimulate demand with integration, power management, and security, and we'll have something.
Don , i agree with what you say, from the viewpoint of mature companies - it's a risky move to self cannibalize and increase price pressure, probably not worth it.
But in many not-so competitive markets , it's usually startups that teach us what's possible. And look there's indie-semi, a startup , who doesn't differentiate on flash/ram - every chip it sells will have 160KB flash/8KB sram - which covers a lot of applications comfortably. And according to their site they even sell $0.20 chips which i understand have the same ram/flash.
I hadn't seen Indie Semiconductor, thanks for passing that on. In simple terms, their model is to miniaturize most of a board into a chip as a custom design. I would note from their website that 20 cent figure is a "typical" scenario with $500k NRE bundled. If you understand exactly what features/package are needed and how many units you are building and how much NRE to collect, dealing for cents from margin is possible. Not disputing their figures, but it's a different biz model than selling a merchant MCU without NRE at 10K pricing and beyond. That said, the Indie biz model is interesting. I ran a similar board level biz for a lot of years - if you can solve the scaling issues, it can work.
Don, they also offer standard products , that go a step closer towards the general mcu.
But let's look at how they do it. They use 2 dies, One digital which has the processor/flash/sram/peripherals , which i believe is done using 55nm, and another analog die , for rf/power/io.
That's why they offer 160KB/8KB sram for free even in their standard products, and this how they could offer freely a variety of peripherals - the only issue is design cost.
And heck according to an eetimes story , they are not even venture funded, all self funded. So it's probably nut hugely expensive to start this.
I think it was Eric that said earlier that in this model, "standard" products start out life as special projects done for one customer with NRE. If that customer allows it - and they often do if there's no proprietary IP and it gets their costs down - after a period of exclusivity the vendor can shop it around for other buyers. That is the way to get self-funding. The first few customers pay NRE which covers engineers and tooling before a single "standard" chip sells.
The idea that the digital and analog dies are at different vendors on different nodes is part of the model as well. The design speed is in recombining blocks quickly. For instance, they could go back to an existing customer in a couple years and say the analog subsystem is OK, we'd like to upgrade the digital subsystem, and it is a relatively low risk change. Or, they could take the analog subsystem to another customer and change out the digital side. The velocity of those two sides is different.
That's a concept we've discussed in "Mobile Unleashed", that the application processor and modem are two different things. Conventional wisdom says integrate for unit cost. At the MCU level, the die cost is important, but the package cost is dominant - if you can swap out one of the two die and keep the package, that is a win for both the vendor and customer (who doesn't have to change PCBs).
Another aspect is who has deep pockets to pay NRE. Many prospective IoT or industrial customers would laugh a vendor out of the building if they said the word NRE, since they are running on a shoestring. But for an automotive or medical customer with very specific requirements and qualification they can't get from other vendors selling standard products, probably not an issue.
The danger in this model lies in sales/marketing. If a bunch of new prospective customers come in all at the same time and want everything customized, engineering has to scale up - and then that engine has to be fed, or you end up with a bunch of engineers sitting around. It's a little better using a fabless model, the supply chain problems are reduced. You need a very selective sales/marketing team that picks the right opportunities and works customer relationships long term instead of chasing quarterly big hits to make a number.
