Consider, if you will, the lowly 8-bit microcontroller. To many designers, it is the epitome of the bad old days-the days when not only performance, but I/O configurations and even processor architecture were dominated by the constraints of a minuscule transistor budget. "Never again" might well be the vow of hardware designers and programmers who have wrestled a slow and helpless 8048 or 6805 MCU into a mildly demanding application.
But 8-bit MCUs refuse to become historical artifacts. Despite the emergence of rich transistor budgets, bringing with them vast on-chip memory, powerful 16-bit and 32-bit processors, and dazzling arrays of smart peripheral blocks, the 8-bitter just keeps clinging to a part of the market. In fact, in a way it may represent the future.
Scaling can do strange things. As process geometries get smaller and smaller, the circuitry on an 8-bit MCU keeps shrinking until even devices with small numbers of I/O pins are pad-limited. And until the 90-nanometer node, smaller meant faster, so the processor cores and memory arrays kept getting speedier. At some point in the process, usually around 350 nm in most companies, the 8-bit MCUs quietly stepped off the Moore's Law curve and stopped shrinking. The pad rings and the practicalities of handling small dice kept them from getting any smaller, and apparently no one needed them to be any faster. So today many 8-bitters are run on very mature processes using fully depreciated fabs.
But a gradual change in control systems design, particularly in areas very rich in sensor input, such as medical research, robotics and process control, may change the future for the 8-bit MCU.
"There is a move in sensor architectures toward distributing the front-end processing of sensor data into the sensor nodes themselves," observed Ata Kahn, director of product innovation at Philips Semiconductors. Kahn said that with growing numbers of sensors gathering increasing amounts of data, it was becoming necessary to shield the system CPU from the blizzard of interrupts and streams of bytes from the sensors. So early processing gets pushed into the sensor. And suddenly, all that blank space that would be inside the pad ring of a 180-nm or even 130-nm 8-bit MCU begins to make sense.
The space makes room for a quite elaborate hardware sensor, including precision analog and lots of storage, virtually free along with the node processor. And the added speed of the advanced processes makes sense too: Sensor data reduction is exactly the sort of flow-through, computationally intensive but moderate-precision job for which a blindingly fast 8-bit core would be, well, blindingly fast.
So a new idea of an 8-bit MCU is emerging. The device would be in an advanced process, and would have a few precision analog inputs and a very few actuator output pins, plus one fast data pipe back to the host system. It would have good precision analog signal capture, conversion and conditioning, modest signal-processing capability and adequate memory, all within the minimum pad ring. Each device type might well be specific to a particular sensor application.
With one more shrink, there's almost room inside the pad ring for a simple RF transceiver. The clunky 8-bit MCU has evolved through the smart sensor into the smart mote, the current darling of distributed control.
Ron Wilson is Semiconductors Editor for EE Times. Feedback and suggestions are welcome at firstname.lastname@example.org