MCU Performance Customers: The Cavalry is Coming Over The Hill

The under the radar, sleepy microcontroller market is about to undergo a rapid transformation the next several years with new entrants and the rise of 32 bit cores that will redefine the parameters for success. This will revive growth and result in new winners and losers. But lots of questions remain.

My first job out of college in 1984 was programming an 8 bit 8051 for a telephone handset. It took months to finish a programming task in assembly language that I thought I could do in a 16 bit microcontroller in a week or so. I begged my boss to allow us to switch. He declined – we couldn’t afford tacking on a couple extra bucks per telephone. Translation: I was underpaid. I swore then that the 8051 would surely be gone in a couple years. Missed that prediction!

It’s still here more than 25 years later. The 8051 along with the other 8 bit controllers are a $5B market and I am now convinced they will never make it to the Smithsonian Museum of History.

What is new is that 32 bit controllers have been on a tear the last two years. They’re finally taking off. This year 32 Bit MCUs should do roughly the same revenue as 8 bit. The magic ASP number for market liftoff is around $1 per chip – unbelievable.

The tragedy of the earthquake and Tsunami that struck Japan highlighted not only how fragile life is but also the world economy. Renesas, the company most severely hit by the earthquake has revenue of $9B that is 40% based on sales to automotive customers. This 40% dimmed the lights in Toyota, Honda and Nissan factories around the world. Think about it – a $9B company, single-sourced, levered into a $1T customer base. Imagine what the entire $15B MCU market leverage is and you see where I am going.

JIT (just in time) manufacturing was exposed and shown to be – at the extreme – very risky. Auto companies will demand 6-9 months instead of 30-60 days of inventory stored around the world. Second sourcing, the curse of the semiconductor industry in the 1970s and 1980s will be asked for but declined. What then are the alternatives that the automakers and others will pursue.

Renesas is the big dog at 30% of the MCU market and they face the supreme challenge of winnowing down the extended number of architectures that resulted from two large mergers. The first was Hitachi and Mitsubishi. More recently NEC. In their efforts to support all legacy products, they risk losing the future. And there is no talk yet of adopting ARM at 32 bits. They appear to feel safe with the current customers but the high end market is pushing for more performance.

ARM is the new love interest of many microcontroller vendors at 32 bit. The argument is that they can be targeted in very low power and high performance markets with different cores. Then there is the common programming platform which customers will appreciate. It can be compelling and seems to be working. Atmel, St Micro, TI, Infineon and others are headed down this path. Microchip has licensed MIPs to attack the 32 bit market and with its leadership in the 8 bit market – they seem be leveraging its loyal customer base for future growth.

The second trend that to me could be more impactful is the fact that Xilinx and Altera have announced plans to enter the market in the next year with a family of FPGAs that include hard core A9 processors running up to 1 Ghz with their associated Caches, hard memory controllers, CAN and Gbit Ethernet controllers. All this with a sea of LUTs and hundreds of GPIO. Ahh – yes but its an FPGA and probably will cost more than an ARM and a Leg (excuse the pun).

This is where I think it is interesting. Xilinx and Altera are focused on 28nm process technology. Much of the 32 bit MCU world is at 130nm or 90nm. By being at least 3 nodes ahead and using hard blocks for the CPU and peripherals, there is a chance that these parts will be smaller in die size than current MCUs and therefore sell at or below price parity. One caveat to this – there is no integrated flash for code store. I suspect they both will include a stacked die arrangement in their product families. Perhaps, though this is an outside chance – Altera and Xilinx will try to be pin compatible with other ARM MCU vendors.

Another aspect to watch is how analog fits into this strategy. Many MCU vendors have seen tight integration of the processor and analog on a single die as the winning formula. However, integrating analog below 90nm is difficult and doesn’t offer Moore’s Law savings. I presume the FPGA vendors are focusing on performance and will partner with leading Analog guys like Analog Devices, Linear Tech for the platform solution.

The auto and industrial markets are the most likely first targets. Automakers are begging for more performance and the 1GHz solutions from Altera and Xilinx are likely to be a leap ahead of Renesas. Plus I would suspect Altera and Xilinx architected their offerings based on input from the smaller set of large auto and industrial customers instead of the thousands of total worldwide MCU customers. Remember 40% of Renesas sales is automotive.

For the end customer – more solutions to choose from. The ingredients are: ARM standard architecture+two vendors at similar pricing+full temperature range (Auto, Industrial, consumer). For Xilinx and Altera it is an interesting new market to pursue. At $5B in size, the 32 bit MCU TAM is larger than their current combined revenue.

A week ago I listened to the Altera earnings conference call. What was intriguing was that John Daane – the CEO mentioned that they had a Japanese customer come in requesting a one time order for their high end Stratix 4 FPGA to replace an ASIC that they couldn’t source due to the Tsunami. The revenue Altera would receive in the coming quarter would be over $15M – significant enough to tell Wall St. I started thinking, the customer had to have redesigned his PCB to support the Stratix 4. But in the future, a customer in a crunch may not have to redesign – just place an order.