Evolution Of The MCU

Microcontrollers are taking on a variety of new and much more complex computing tasks, evolving from standalone chips to more highly integrated devices that can rival complex microprocessors.

Microcontroller units (MCUs) are being designed into everything from assisted and autonomous driving to smart cards. They often are the central processing elements for a slew of connected devices that increasingly define the IoT. In fact, there is debate over whether the IoT would even be successful without MCUs.

Fig. 1: MCU demand is showing healthy growth, but total market revenues are growing at a much slower pace. Source: Semico Research

This pervasiveness comes at a price, though. While market opportunities are expanding, average selling prices for MCUs are plummeting. It now requires much higher volumes to recoup development costs, and without some big changes it may be difficult for some companies to continue profiting in the MCU space.

Fig. 2: The biggest reported increases in revenue for MCUs was from consolidation. Source: IC Insights

Microcontrollers come in three main flavors, 8-, 16- and 32-bit. There are still some 4-bit microcontrollers in use, but many 4-bit customers either have moved to 8-bit or plan to move in the future. The next step up in functionality would be 16-bit MCUs, which are more expensive. But prices have fallen so fast for 32-bit MCUs that the majority of customers that were considering 16-bit upgrades are now moving to the more advanced chips.

“This year, 32-bit ASPs dropped 15%,” said Rob Lineback, senior research market analyst at IC Insights. “We’re seeing 32-bit prices falling 7% per year (CAGR). The 16-bit market is minus 2% and 8/4-bit is minus 2%. It will reach a point where 32-bit microcontrollers are priced lower than 16-bit microcontrollers. This is pure competition and pricing pressure. IoT is part of the driving force for this.”

Joanne Itow, Semico Research’s managing director for manufacturing, is seeing a similar trend. “ASPs in all logic markets, except for DRAM and NAND, aren’t doing well right now. The only big winner in the microcontroller market is ARM, because they’ve got the cores, the IP for low power, and the security. Companies can develop it all themselves, but it’s easier to use what ARM has.”

Same issues, different strategies
There are two conflicting trends at work here—ASPs are falling and complexity is rising. That puts much more pressure on MCU makers to find efficiencies in their processes and methodologies.

One approach is to take advantage of what is already available inside these devices, but which many companies currently ignore. “If you look at TrustZone, those capabilities are largely not being utilized,” said Andrew Caples, senior product line manager for the Embedded Software Division at Mentor, a Siemens business. “You can add reliability and create memory partitions so that you can read only or read and write, but that’s being underutilized. Power management is another feature, which in some cases may be the determining factor for whether a product does well or fails because it affects battery life. Some of these MCUs have more than 16 low-power states, but to transition from one power state to the next is complex. There needs to be far more thought put into the software to take advantage of these features.”

Rising complexity has been inducing MCU makers to move to the next process nodes, where more memory, connectivity and processing can be crammed into the same space. This is applied to a different market, and for 32-bit MCUs the leading-edge node today is 40nm. Companies are working on 32/28nm versions, as well.

“The problem is that microcontroller companies develop dozens or even hundreds of SKUs (stock-keeping units), in part because the pins assigned to serial I/Os vary,” said , CEO of Flex Logix. “Some are SPI (serial peripheral interface), some are UART (universal asynchronous receiver/transmitter). Or they provide all the hardware and bond out differently. But at 40nm, mask costs are going up, so dozens of variations cost a lot of money. It takes a certain number of look-up tables to program a serial I/O.”

One way around that is to add flexibility into the microcontroller itself with an embedded FPGA, so these devices can be programmed for a variety of markets rather than developing a new MCU for each application.

A third approach is to be more efficient with verification, reducing the amount of time needed on the back end of the MCU design flow.

“This is why Portable Stimulus is so interesting,” said Frank Schirrmeister, senior group director for product management and marketing for emulation, FPGA-based prototyping and hardware/software enablement at Cadence. “It makes it easier to understand. Some of these MCUs are growing up to be more system-like, so what some of the big microcontroller companies are doing is selling those designs with customized software development around it. That can be used to validate it.”

One device, many uses
All of these approaches help turn an MCU into an off-the-shelf device that can be quickly customized for vertical market segments or specific applications. But the whole process needs to be simplified even further.

“In EDA you’re more tolerant of some quirks in the tools,” said , senior director of market development at Arm. “If you look at the market for microcontrollers, it’s different because the software folks want to download debuggers and compilers and make it work. The end product for them is not the chip. It’s the system, and they need something that will solve their problems effectively. What you’re providing is a shrink-wrapped device with lots of features and options, so you need to satisfy everyone from the most advanced users to the one-off product developed by a hobbyist. The same product needs to work for all of them.”

The key is to strike a balance between time to market and cost on one side, and optimization on the other. That balance may shift significantly, depending upon the application.

“There are a number of different approaches,” said Semico’s Itow. “Some companies design for specific applications, like the IIoT. Some believe a generic approach is the right one and then customize around it. But what’s clear is there are a variety of opportunities. So even though we’re seeing some consolidation, there is enough opportunity that we will see new companies entering this market.”

The real volume is in the IoT and the consumer markets, and MCUs are tailor-made for these segments. They contain a mix of memory, processing, security and communications technology, but less of everything than a powerful SoC.

“In the past, a lot of microcontrollers ended up in the industrial and automotive markets, which had longer sales cycles,” Neifert said. “Now, we’re seeing more of them in the consumer space, where content needs to be the latest and greatest. With an increasing number of these devices being marketed to the consumer, you need rapid turnaround, meaning the same microcontroller may be applied in 10 different ways, and a lot of that depends on the software.”

Some of those applications are for systems that either never existed before, or in which a microcontroller was more of a generic add-in than a key part of the architecture.

“There are higher-performance cores and clocking that are normally not associated with microprocessors,” said Bill Hutchings, senior product marketing manager at Microchip Technology. “So if you look at a sensor hub, there usually is a pre-processor, which typically is a mid-tier microcontroller.”

Redefining MCUs
In years past, what differentiated one microcontroller from another was functionality. So a 32-bit MCU offered significantly more functionality than an 8-bit MCU, and it was priced accordingly. But as companies migrate upward, they are now working with devices that are on par with the lower end of the microprocessor line.

Historically, though, there were still significant differences. A microprocessor uses a combination of internal and external memory, while an MCU traditionally relied exclusively on internal memory. That is beginning to change, with some MCUs providing capabilities to connect to DRAM or other types of external memory.

“If you go back 15 years, you could take 100 engineers and put a microcontroller and a microprocessor in front of them and they could pick which was which,” said Steven Woo, distinguished inventor at Rambus. “Today if you took those same 100 engineers, you would have a lot of debate. Moore’s Law is part of what’s causing this blurring. There are more transistors on a die, and there’s a lot more that you can do with those transistors.”

Along with greater real estate is the ability to put more memory on-chip and off-chip. MCUs typically use a combination of DDR2, as well as flash memory. But because of increased density everywhere, memory sizes for DDR2 have increased to as high as 2 megabytes, with up and as much as 2MB of embedded flash.

“These devices have had the ability to connect to external memory for at least six or seven years on the 32-bit product line, but not many of these devices have taken advantage of that,” said Stuart McLaren, STMicroelectronics’ microcontroller product manager for the Americas. “More recently we’ve seen them using [a system packet interface], so there is more NVM externally for them to store data and code. The key change is more performance and more features. There are more applications being connected. And as they becoming connected, you have a node talking to a gateway or maybe a cloud. That requires at least a simple microcontroller, which can collect a bit of sensor data and aggregate it.”

Microcontrollers also are beginning to move upstream in the cloud.

“There are a lot of cloud services running on gateways, and they are doing more advanced analytical work,” said McLaren. “We see external memory for graphics, typically a 1- or 2-frame buffer where you render information from that frame buffer and then refresh. We also see a lot of microcontrollers going into three areas of the IoT—home and city, smart industry, and smart things. Every application has a need for processing and security. They also need to interface with the real work and they need some sort of connectivity, which in many cases is RF, but it also could include Bluetooth Low Energy or some other near-field processing. And they need to manage power, which could be anything from wearables up to industrial. The MCU is the core of the IoT.”

They also are being used for new applications, such as asset tracking, according to Dipesh Patel, vice president for engineering for physical IP at Arm. “With 32-bit, you can track a part through to its journey, and you can make it all secure, which you could never do with 8 bits. MCUs are getting more and more complex. At a simple level, you can store, process and communicate data. But now you can do more because there is more memory.”

He noted there is a move to add on-chip flash memory, as well, particularly as some of the more advanced designs migrate to 40nm and eventually to 28nm. Today, most MCUs use older processes, some as high as 350nm, although some of the more advanced IoT designs are being developed at 65 and 40nm.

Mass customization vs. niche
One of the key drivers for MCUs has always been cost, although in the past decade they have been utilized as often for their power profile, which is why it’s rare to find systems without microcontrollers. But increasingly, they also are being designed for tasks for which systems vendors never would have considered an MCU a few years ago.

“We’re seeing it used as a companion chip for security alongside fully integrated microprocessors,” said Jeanette Wilson, product marketing manager at Microchip. “You may need to do basic authentication, and a microcontroller can be used for hardened key storage. This requires a handshake with an ECC (elliptic-curve cryptography) SoC, and you can put this in an anti-tamper package or use it as an entropy source. You also can add encryption/decryption, with is another level of security capability, with monotonic counters to prevent replay. This is frequently done in software, but you can save 8,000 to 12,000 lines of code if you do it in hardware, and it’s faster from an execution standpoint.”

MCUs also are showing up on expansion boards and alongside of sensors, which can measure everything from motion to temperature.

“We’re running these as high as 400MHz, which you would only see in an MPU in the past,” said ST’s McLaren. “And we’re seeing cases where a microcontroller itself can have multiple microcontrollers connected to it, with maybe an API between the main one and the others.”

Conclusion
Trying to differentiate different kinds of logic devices is becoming more difficult, and it may become tougher still as advanced packaging continues to make inroads into more designs.

“In the 1980s, an MCU was a system on a chip, or a computer on a chip as it was sometimes called,” said IC Insights’ Lineback. “Eventually, WSTS (World Semiconductor Trade Statistics), which is the group that defines these categories, will draw the lines differently.”

In the meantime, MCU vendors will do everything they can to boost the value of their devices and stall the price erosion that has turned this into a challenging technology sector. Whether that means adding more flexibility into devices, pruning the numbers of devices being developed today, or changing the way these devices are designed and verified isn’t clear. But MCUs are only growing in importance, so these issues will need to be solved.

Related Stories
The Mightier Microcontroller
MCUs will play an increasingly critical role in cars and IoT devices, but they’re also becoming much more difficult to design.
MCU Sales Up In 2017 And 2018
Where are all the microcontrollers going?
New Embedded Memories Ahead
Options grow as new wave of MCUs demand more capable NVM.
IoT Myth Busting
How cost-sensitive are IoT edge devices, what are the real drivers for this industry, and what is the impact on EDA and IP?