From one perspective, the applicability of 16-bit MCUs has become limited, squeezed by capable and low-power 8-bit MCUs on the low end, and 32-bit MCUs that are quickly dropping in price at the high end. Indeed some MCU manufacturers are focusing their product development entirely on 8- and 32-bit products. Time will tell if that strategy is flawed, because the latest 16-bit MCUs offer near parity with the 8-bit competition in terms of system-level power consumption, along with a significant performance advantage. It is possible we could see an expansion of the application segment served by 16-bit devices especially with the increasing need for connectivity in most electronic systems. Let's examine some of the latest 16-bit technology developments relative to the 8- and 32-bit competition.
First we should address the general factors that influence the choice of 8-, 16-, or 32-bit devices. At the low end, the decision comes down most often to low cost and low power consumption. Generally, 8-bit devices win out in those areas. But, as we will cover here, there are 16-bit options that can compete with some 8-bit MCUs in terms of both power and cost.
There are certainly other factors that can come into play. For example, we covered the 8-bit 6- and 8-pin MCU segment in a TechZone article last year (Miniature Six-Pin MCUs Offer Surprising Functionality). If minimal footprint on your printed circuit board is a prime decision factor, you may not find 16-bit devices that can match the smallest 8-bit devices.
Decisions between 16- and 32-bit MCUs most often come down to the combination of performance and peripherals that can handle the task at hand balanced against power consumption and cost. Microchip product marketing manager Alexis Alcott said, "It's a lot less about bits, and more about the other requirements."
True 16-bit cores maintain a code-density advantage over 32-bit MCUs, meaning an application generally fits in less memory. That advantage equates to lower power and cost. But 32-bit architectures such as the ARM Cortex family seek to support smaller code size requirements by offering a mix of 16- and 32-bit instructions. ARM calls the capability Thumb-2 and most MCU vendors support the optimal instruction set in ARM Cortex-M0/M3/M4-based MCUs.
The 32-bit MCU class generally offers higher clock rates and better performance. But 16-bit offerings are quite capable, as we will cover shortly. Moreover, vendors are adding math capability such as DSP-centric multiply-accumulate (MAC) units and floating-point units (FPUs) in hardware on 16-bit MCUs, reducing or eliminating any performance advantage for low-end 32-bit MCUs.
System power consumption
Since power consumption is a very important consideration, let's start our examination of 16-bit MCUs focused on that characteristic. The 8-bit offerings offer a baseline advantage for several reasons. Manufacturers tune the process technology used in manufacturing 8-bit devices to low power. Transistors are optimized for low leakage rather than fast switching times. So when you peruse data sheets, you will see a consistent advantage for 8-bit MCUs in terms of active power consumption.
But the story runs deeper. MCU designers have followed the lead of mainstream microprocessors in developing a robust set of low-power operating modes that can significantly impact the actual power used at the system level in a real application. In many applications, the bulk of the circuitry in an MCU can remain in a low-power mode most of the time, awakening only based on interrupts.
Texas Instruments (TI) has been a pioneer in the area of targeting traditional 8-bit applications with the 16-bit core implemented in the MSP430 family, and power management is a key to the success of the product. TI points out applications such as smart meters, portable medical devices, and datalogging systems where system requirements allow the MCU to enter sleep mode and awaken based on sporadic or periodic interrupts generated by timers or external events.
Low-power modes and clock
The MSP430 includes a total of eight operating modes. The granularity of the scheme is enabled by a flexible clocking system that is illustrated in Figure 1. The main CPU core is driven by the MCLK (main clock) generated by an integrated, DCO (digitally-controlled oscillator) that supports rates that range from 100 kHz to 25 MHz.
Figure 1: Three clocks enable a robust set of low-power operating modes in the TI MSP430 16-bit MCU family.
The auxiliary clock (ACLK) drives peripherals at lower speeds and can be derived from a low-power on-chip oscillator or an external crystal with a max rate of 32 kHz. There is also a third clock, called sub-main clock (SMCLK), not included in the figure that can use the DCO output to drive peripherals requiring a faster clock.
The application code must leverage the low-power modes to minimize system power. The first step down kills the MCLK but maintains the other two clocks. One step further limits the SMCLK to lower-speed operation and the following step kills that clock. Ultimately the ACLK is killed meaning that the MCU doesn't have timers active to generate an interrupt. Ultimately only an external interrupt is able to awaken the MCU and in the lowest-power mode the RAM contents aren't retained.
The low-power modes deliver impressive results. MSP430 devices can retain RAM contents consuming only 0.1 µA. The real-time clock remains active in a mode consuming less than 1 µA. Active power consumption is determined by the MCLK speed and TI specifies that usage at less than 100 µA/MHz. You can clearly see the significant savings afforded by the low-power modes.
Microchip XLP technology
Microchip is another vendor of 16-bit MCUs that supports an extensive set of low-power modes. The company uses the brand XLP (extreme low power) for its lowest-power devices across the 8-, 16-, and 32-bit families. XLP implies a deep-sleep, low-power mode that's significantly lower than 0.1 µA across the portfolio. Of course, some parts deliver much lower values. For example, the 8-bit PIC18F47J13 uses only 9 nA in the lowest-power mode. XLP also implies current consumption below 0.8 µA with the watch-dog timer and real-time clock active.
Microchip's Alcott said that the company achieves low-power using different techniques across the portfolio. The focus in 8-bit devices is on the manufacturing process. Achieving XLP-level results in 16- and 32-bit MCUs is more focused on low-power modes, according to Alcott.
Let's consider a couple of examples of XLP products in the 16-bit PIC24 MCU family. The 1.8 V version of the PIC24F16KA102 MCU features deep-sleep current draw as low as 20 nA. Meanwhile the newly-announced PIC24FJ128GA310 MCU (Figure 2) introduced some new low-power-mode enhancements that lower the minimum current draw to 10 nA in deep-sleep mode. Clearly the 16-bit devices are competitive with the 8-bit devices in minimum power levels.
Figure 2: Microchip offers a deep-sleep mode on the new 16-bit PIC24FJ128GA310 MCU that retains the contents of 8 Kbytes of RAM using 330 nA.
The story is similar in the lowest-power mode with RAM retention, although the RAM retention spec depends both on the low-power technology and on the amount of memory integrated on chip. Microchip touts a new deep-sleep with RAM retention spec of 330 nA for the PIC24FJ128GA310 MCU that integrates 8 Kbytes of RAM. The PIC24F16KA102 MCU can retain RAM at only 25 nA, but only integrates 1,536 bytes of RAM. The 8-bit PIC18F47J13 retains RAM at 200 nA, and integrates 3,800 bytes.
Also offering a strong 16-bit portfolio, Renesas has several product families that feature low-power operating modes including the RL78 that just came to the market last year. Renesas takes a slightly different approach to low-power, focusing on ways to allow some key peripherals to operate in low-power modes while minimizing the need to awaken the CPU and enter full active mode.
The RL78 has four primary operating modes, as shown in Figure 3. The lowest-power mode is the Stop mode in which RAM contents are retained. Different family members range in current draw from 0.2 to 0.6 µA.
It's the Snooze mode, however, that may prove most useful in minimizing system power. The A/D converter and serial-communication functions remain active in Snooze mode. Moreover, there are integrated capabilities in the converter to compare samples against preset limits. A system could monitor a critical process via a temperature sensor or accelerometer using the A/D converter. As long as the results stay within the preset range, the CPU remains asleep. When a value falls outside the prescribed range, the converter awakens the CPU with an interrupt.
RL78 MCUs typically consume around 5 mA in active mode at full-speed, 32-MHz operation. Running the data converter in Snooze mode consumes 0.5 mA. Renesas contributed an article to Digi-Key's microcontroller resource library that more fully explains the RL78 architecture and Snooze-mode operation.
Figure 3: A Snooze mode on the Renesas RL78 allows an A/D converter to remain active while the CPU sleeps and a comparator function checks converter samples for in-range values.
A couple of other facts about the RL78 are worth noting. The low-power features of the architecture apply across the entire MCU family. There are no select low-power offerings. Nelson Quintana, senior marketing manager at Renesas, pointed out that the RL78 family already spans devices that range from 20 to 128 pins, 2 to 512 Kbytes of flash, and 0.25 to 32 Kbytes of RAM.
Spurring the 16-bit decision
Having covered low-power and established that 16-bit MCUs are in the ball game with 8-bit MCUs, let's consider what might prompt the choice of the 16-bit option. Memory seems the most likely reason that design teams will choose a 16-bit MCU.
Renesas' Quintana said, "There's not enough code space in 8-bit MCUs for many applications." He noted that many 8-bit families top out at 128 Kbytes of flash. For example, Quintana identified the need for safety compliance in many embedded products and that requires additional code storage to host the safety routines. He also said that medical applications and any system with connectivity need storage beyond the capability of an 8-bit MCU.
Microchip's Alcott made a similar point about the available RAM that is generally limited to 2 to 4 Kbytes on 8-bit MCUs. Alcott said anything with a protocol stack, such as TCP/IP or ZigBee, can be implemented more effectively on a 16-bit MCU because such stacks require temporary storage of many data values.
Ease of programming
The available memory in 8-bit MCUs can also impact the ease of programming. Alcott said, "Many customers choose the 16-bit PIC24 family because it is very efficient for C code." Microchip supports C for its 8-bit product as well, but the limited memory leads to hand-optimized assembly language in many instances.
The C-language trend could certainly help broaden the 16-bit application segment. Alcott said that many new graduates of engineering and computer-science schools today don't learn assembly-language programming.
Having covered many of the decision points that line the 8- and 16-bit border, let's turn to the border where 16-bit MCUs compete with 32-bit alternatives. The latest 16-bit families certainly overlap with 32-bit MCUs in terms of performance, and that's a primary decision factor for many designers.
Benchmarks are a tricky topic, but they are the only viable way to offer comparisons here. Clearly clock speed matters, but the different microarchitectures, used in various MCUs, mean that clock speed is not a good first-order indicator of system performance. So let's discuss the Dhrystone MIPS (DMIPS) rating for a couple of processors.
ARM and its Cortex partners have said that the Cortex-M0 that is the low end of the 32-bit Cortex family delivers 0.9 DMIPS/MHz. Renesas says that the RL78 MCUs deliver 41 DMIPS at 33 MHz, or about 1.3 DMIPS/MHz. Companies such as NXP, however, already have Cortex-M0-based MCUs, such as the LPC1100 family, that run at 50 MHz and the maximum clock speed on 32-bit devices will continue to escalate. Still, 16-bit MCUs are a viable option to a 32-bit device in many applications.
Renesas also offers some hardware math functions that can boost performance in compute- and DSP-centric applications. The MCU includes a MAC that can handle a 16x16-bit multiply in a single cycle and the full MAC operation in two cycles. The MCU also integrates a barrel shifter.
Microchip's Alcott said that comparing candidate MCUs in the actual application is the best way to ensure that a selection matches your performance requirements. Even though the PIC24 16-bit family and the PIC32 32-bit family use decidedly different instruction sets, Microchip provides the same IDE (integrated development environment) and compatible C compilers for both families. For the most part, the same code can be compiled for each in a head-to-head test.
Peripherals and price
The decision points that we haven't yet covered may prove the most important in your particular project. In particular, the peripheral mix on an MCU and the price are important characteristics.
Faster peripherals can lead a design team to a 16-bit MCU as opposed to the 8-bit option according to Alcott. She also noted that unique peripherals often drive the 16-bit choice citing the graphics controller on the 16-bit PIC24FJ256DA210 MCU. The controller drives a relatively-small QVGA graphics display, but that added integration is key in applications that require and/or can afford the option.
Price is another almost intangible element. You simply have to look at pricing of specific MCUs relative to your work. Renesas' Quintana said, "Not long ago, there was nothing under $2.00 in the 32-bit space. We're now seeing sub $1."
“There are many 16-bit MCUs prices at less than $0.50,” Alcott added, "The entire industry is seeing a pretty intensive price crunch."
You just may be comparing pennies when you consider MCU options these days. But pennies add up if you manufacture a product in high volumes.
What is clear is that you should certainly consider 16-bit MCUs when facing a design project even if some manufacturers have decided to ignore the segment. As Microchip's Alcott noted, "The 16-bit MCU is one of our fastest-growing sectors, especially for new customers."
Indeed, manufacturers of the latest 16-bit MCUs promise to capture some design wins on both ends of the application spectrum. The 8-bit segment appears especially vulnerable. As manufacturers continue to innovate in low-power modes, it could be that 16-bit MCUs yield lower system power than 8-bit MCUs even if the 16-bit devices have higher active power consumption. The 16-bit MCUs can spend shorter periods in active mode because they offer better performance and can finish the task at hand and re-enter sleep mode more quickly.