Tensilica's Steve Leibson makes a strong case for the end of 16-bit processors, as he says 32-bit devices offer a clear advantage over their smaller brethren
Like many 19th and 20th-century technological artifacts — the steam locomotive, 8-track and compact cassette audio tapes, black-and-white television, the slide rule, and 4-bit microprocessors — 16-bit processors have reached the end of their useful life. Yes, you can still buy them, but you no longer need them. In fact, they now hamper your ability to design systems in a timely fashion. They dull your design team’s competitive edge.
The first single-chip microprocessor was a 4-bit device — Intel’s 4004 introduced in November, 1971 (about 2300 transistors). Intel essentially superseded its 4004 just five months later with the 8-bit 8008 microprocessor (3500 transistors). However, 4-bit microprocessors survived for more than two decades, mostly as low-cost microcontrollers used in low-end consumer products.
The advance of Moores’ law has completely outmoded 4-bit microcontrollers in favour of 8-bit processors that are far easier to program, which shortens time to market.
Billed as the next step up from 8-bit processors, the 16-bit processors that first appeared in the late 1970s promised better system performance for a little more silicon. However, the 16-bit processors’ address spaces were still mostly limited to the same 64kbyte available in the 8-bit programming models (although some 16-bit processors, including Intel’s 8086, used memory-segmentation schemes to expand the address range, at the expense of additional code complexity).
 |
Steve Leibson doesn't think much
of 16-bit processors
|
The first microprocessor with a 32-bit programming model, the Motorola 68000, also appeared in the late 1970s. At first, 32-bit microprocessors were used only in high-end workstations while 16-bit processors were used in PCs and, increasingly, in embedded designs. However, as IC lithography improved, 32-bit processors shrank in size and cost so that 32-bit processors are now routinely used in products sold across all commercial, consumer, industrial, medical, and military and markets.
The “right” microprocessor is the one that gets your product to market on time, on budget, with all the desired features, and for the lowest cost. For years, 16-bit processors have had the edge over 32-bit processors with respect to cost — for two reasons. The first is the cost of memory. Processors with 16-bit ISAs (instruction-set architectures) tend to require less program memory than processors with 32-bit ISAs and memory costs money. The second reason is the cost of the actual processor: 16-bit processors are physically somewhat smaller than 32-bit processors and therefore have lower silicon costs.
These 16-bit processors’ cost advantages are counterbalanced by the compromises of a 16-bit ISA, namely: limited memory and register addressing. Due to these limitations, 16-bit processors are restricted to a few registers and to 64kbyte addressing (128kbyte for architectures with separate instruction and data spaces) while 32-bit processors have large register files and 4Gbyte address spaces. Using address-restricted 16-bit architectures becomes increasingly difficult in modern products that frequently deal with “rich” media such as mp3 music files, jpeg images, and mpeg video.
In addition, embedded coding styles have largely shifted from assembly language to high-level language compilation and compilers are far more efficient when given large register files. Nearly all modern media-processing algorithms are now written in C or C++ so there has been substantial research on optimising compilers to produce tight, efficient 32-bit code.
All major 32-bit processors now have 16-bit subsets of their instruction sets to reduce the instruction-memory footprint. For example, Tensilica’s 32-bit processors employ a 24/16-bit instruction set, which results in very small instruction-memory footprints.
These advances in 32-bit processor architecture blunt the 16-bit processors’ two cost advantages while retaining the 32-bit processors’ advantages of large register files and 4Gbyte address spaces.
Further, silicon costs for 32-bit processors have not remained constant. With rapidly advancing semiconductor process technology, the 130nm node is now the mainstream process for system-on-chip manufacturing and some full-featured 32-bit processors require only 0.5mm² of 130nm silicon while 64kbyte of RAM consume nearly 2mm².
Consequently, 32-bit processors are now quite small and inexpensive. Despite claims that 16-bit processors require only half as much silicon as 32-bit processors, the real number is more like a third less silicon, which is remarkably similar to the transistor-count ratio between Intel’s 4-bit 4004 and 8-bit 8008 microprocessors. Thus the savings from using a 16-bit processor fabricated in 130nm silicon can be as little as 0.2mm², which is vanishingly small relative to overall system costs. The difference is even smaller for more advanced IC lithographies.
There are no architectural features in 16-bit processors that 32-bit processors cannot offer to programmers through more compiler-efficient ISAs. With the cost differentials vanishing and their tremendous architectural superiority, 32-bit processors are now rapidly displacing 16-bit processors in standard microcontrollers and in SOC design. The 16-bit processor has served the industry very well, but its time is now past. Long live 32 bits.
Steve Leibson is strategic marketing manager at Tensilica