PHILIPSSEMICONDUCTORS
`
`solid state
`
`[1] Microcontrollers are self-contained ICs designed to perform a specific function by themselves.
`This 8-bit CMOS microcontroller chip has 256B of RAM, 8KB of ROM, and another 8KB of elec-
`trically programmable ROM. It runs at 16 MHz.
`
`Workhorses of
`
`Today’s microcontrollers are performing better than ever through
`their use of high-level languages and multitasking techniques
`
`ACLOSER LOOK AT THE MACHINES AND APPLIANCES
`
`that serve us well in our daily lives almost always
`reveals a microcontroller within. Controllers are em-
`bedded in cordless and portable telephones, point-
`of-sale retail electronic cash registers, scanners of all kinds, secu-
`rity systems, automobiles and gas pumps, automated tellers,
`computers, and compact disks and disk drives, not to mention
`phone-answering, fax, vending, and washing machines. In fact,
`electric lights are almost the only electrically powered devices
`that do not use microcontrollers, and even here things are chang-
`ing with the welcome advent of power-saving and quick-starting
`intelligent ballast fluorescent lamps.
`Still, although microcontrollers and
`microprocessors share many architec-
`tural features, they differ in important
`respects [Table 1]. A microcontroller is
`
`ATA R. KHAN
`Philips
`Semiconductors
`
`generally a one-chip integrated system meant to be embedded
`in a single application; so it is likely to include all the peripher-
`al features—program and data memory, ports, and related sub-
`systems—needed for the computer aspect of the application
`[Fig. 1]. By contrast, a microprocessor drives a general-purpose
`computer whose ultimate application is not known to the system
`designers.
`Microcontrollers may not enjoy the publicity of their more
`glamorous cousins. Still, they are a key part of the computing
`landscape, as their sales and ubiquity prove [Fig. 2].
`While certain sectors of the microcontroller market—for ex-
`ample, the large 4-bit segment—are essentially stagnant, the
`high end of the market is growing rapidly. Designs incorporat-
`ing 16-bit microcontrollers continue to proliferate, and even 32-
`bit embedded control applications are beginning to appear in
`the industry [Fig. 3].
`
`36
`
`0018-9235/96/$5.00©1996 IEEE
`
`IEEE SPECTRUM OCTOBER 1996
`
`Authorized licensed use limited to: Sterne Kessler Goldstein Fox. Downloaded on December 05,2022 at 20:39:11 UTC from IEEE Xplore. Restrictions apply.
`
`VWGoA EX1019
`U.S. Patent No. 9,955,551
`
`

`

`The size of today’s microcontroller market is fostering intense
`competition that helps spur technical innovation, benefiting users
`and designers of microcontroller systems by forcing down prices
`and creating a wide range of choices. In 1994, the top 10 micro-
`controller vendors (in descending order, based on revenues) were
`Motorola, NEC, Mitsubishi, Hitachi, Intel, Matsushita, Philips,
`SGS-Thomson, Toshiba, and Fujitsu. These familiar names have
`been joined by many third-party suppliers that provide a large
`assortment of development tools, real-time operating systems,
`and other products used to develop and implement the many
`applications microcontrollers have spawned.
`
`Trends in system development
`Low-cost consumer items served by 4-bit controllers include
`microwave ovens, shavers, toasters, white goods, and tape play-
`ers, to name just a few. The 8-bit market includes such embed-
`ded-control applications as television sets, disk drives, car radios,
`and auto body electronics, as well as personal computer periph-
`eral systems (including printers, joysticks, modems, and mice). As
`compared with 8-bit controllers, 4-bit chips are ill-suited to run-
`ning programs written in high-level languages, almost all of
`which are byte-oriented. So a 4-bit chip would have to execute
`multiple operations to do the simplest things. In addition, 4-bit
`controllers have very little compiler support. Programming must
`therefore be done in assembly languages, lengthening develop-
`ment cycles and creating a maze of problems for maintaining and
`upgrading code.
`Meanwhile, 16-bit controllers are generally deployed in disk
`drives, automotive engine controls, and industrial control appli-
`
`In fact, for 16- and 32-bit applications, systems are now being
`developed exclusively with high-level languages. When opti-
`mized for microcontrollers, they yield efficient compilation pro-
`grams that speed development and time-to-market. For maxi-
`mum code density, the best course is to write assembly code by
`hand—assuming the designer has the advanced and increasing-
`ly rare expertise required to do so. But any loss of code density
`that occurs when code is written in a high-level language is more
`than offset by improvements in the product development time.
`
`Real-time kernels…
`As programs grow larger in the real-time environments of
`many embedded control applications, the microcontroller has to
`coordinate demands from various parts of the system and sched-
`ule and service them. These duties call for multitasking capabil-
`ities and hence for real-time kernels (RTKs).
`The kernels lay the groundwork for the required efficiency
`with their scheduling and orderly task execution facilities—tasks
`must be prioritized and deterministic, with known latency and
`so forth. In more detail, RTKs include mechanisms for starting
`and completing tasks, for switching between them in response
`to various priorities, and for facilitating communication among
`tasks for such purposes as synchronization, resource sharing, and
`sequencing.
`
`…and operating systems
`Microcontroller operating systems that are extensions of
`RTKs, providing real-time control constructs, are called real-
`time operating systems. They combine the task-control, real-
`
`the electronic era
`
`cations. And 32-bit devices are becoming more common in
`three principal areas: communications boards (such as network
`bridges and routers), laser and ink-jet printers, and x-terminals.
`Video games represent yet another opportunity for 32-bit
`embedded controllers, as well as for 64-bit ones.
`With such widespread application, microcontrollers are sub-
`ject to the same pressures that shape all segments of the elec-
`tronics marketplace: how to achieve higher levels of performance
`and integration while not costing themselves out of a highly
`charged business environment where every penny counts.
`System complexity and performance requirements have kept
`on growing. Meanwhile, product design cycles have shriveled to
`6 to 12 months, compelling system suppliers to improve time-to-
`market and exploit ever-dwindling windows of opportunity. As a
`result abstraction and code reusability have become critical. But
`because these features are not adequately addressed by assembly
`language development methods, high-level languages, real-time
`kernels, and real-time operating systems are proliferating.
`
`High-level languages
`Once dismissed as inefficient for 8-bit microcontroller designs,
`such high-level languages as C are now ubiquitous—and are
`even being used in systems with as little as 2KB of ROM on chip.
`
`1. A microprocessor and
`microcontroller compared
`Intel 486
`IDT R36100
`
`Feature
`
`Central processing unit
`
`32 bits
`
`Cache memory
`
`Prefetch queue
`
`Floating-point unit
`
`Memory management unit
`
`Communication controller
`
`Direct memory access controller
`
`Timers
`
`Memory controller
`
`P1284 parallel port
`
`I/O ports
`
`8KB unified
`
`4KB inst.,
`1KB data
`
`✔
`
`✔
`
`✔
`
`—
`
`—
`
`—
`
`—
`
`—
`
`—
`
`—
`
`—
`
`—
`
`✔
`
`✔
`
`✔
`
`✔
`
`✔
`
`✔
`
`KHAN — WORKHORSES OF THE ELECTRONIC ERA
`
`37
`
`Authorized licensed use limited to: Sterne Kessler Goldstein Fox. Downloaded on December 05,2022 at 20:39:11 UTC from IEEE Xplore. Restrictions apply.
`
`

`

`time–driven facilities of an RTK with the standard features
`and services of an ordinary operating system, including the
`integration of languages, debugging environments, utili-
`ties, and file management.
`While quite common in 16- and 32-bit microcontrol-
`lers, which are dominated by high-level languages, real-
`time operating systems and kernels have also been created
`for 8-bit devices. These real-time operating systems supply
`a range of program control, task-switching, and resource-
`allocation functions for complex applications where it is
`not feasible to write code by hand. Designers have found
`it much simpler and more efficient to use the optimized
`real-time kernels and operating systems available from
`today’s wide range of third-party vendors.
`Real-time kernels, executives, and operating systems
`must provide quite a few services. One is multitasking, for
`more than one task must be executed on one and the same
`system without interfering with one another. They may
`proceed to completion or be interrupted while executing,
`in which case their state must be saved so it can be
`resumed precisely. A task that fails to execute properly—
`whether because of bad programming, unexpected inputs,
`system glitches, or other factors—should not cause other
`tasks or the system itself to crash.
`Second, the executive must be able to preempt tasks
`that are running so that more urgent ones may execute. If
`required, it must be possible to execute tasks in a specified,
`sequential manner.
`Third, it may be necessary to support real-time clock-
`based operation, because some events are absolutely or
`relatively time-dependent: they must occur within, be-
`fore, or after some specified time or interval.
`Then, too, communications and resources must be syn-
`chronized. Tasks must communicate with one another in
`an orderly and predictable fashion within the context of a
`protected environment. In addition, they must pass re-
`sources back and forth, without having those resources
`overwritten by intervening tasks. Resource sharing must
`occur in such a way that tasks receive the proper attention
`without monopolizing system resources. Such classic
`large-machine operating system techniques as semaphores
`and queues fulfill these functions.
`Finally, a kernel or executive has to provide for efficient
`switching among tasks. Efficiency can be defined as appro-
`priate use of time or resources, or both, achieved through
`
`tradeoffs permitting users to select the degree of context
`saved. Time is exchanged for the amount of machine and
`task state that must be saved and later restored.
`A real-time kernel should use system and chip resources
`economically. Embedded control is highly cost-sensitive,
`and memory size ought to increase minimally if at all. A
`real-time executive that makes large demands on memory
`may in fact reduce the performance of the system and add
`to its cost.
`
`Trends in development tools…
`The development tools and environments that come
`along with the silicon have become integral to today’s
`microcontroller architectures. As clock speeds increase
`and real-time development becomes a must, it has got-
`ten more and more difficult to emulate and debug the
`target system. This simply cannot be relegated to an
`afterthought.
`Indeed, the emulation and debugging of circuits must
`be built into the microcontroller design. Some aspects of
`these features (such as breakpoint/trace instructions and
`special cache instructions) may be visible at the archit-
`ectural or instruction-set level, while other aspects are
`accessed through special emulation/debug modes. With
`various degrees of effectiveness, every advanced micro-
`controller today provides on-chip support for emulation
`and debugging.
`As the use of assembly language has tapered off, tool
`suites have pushed to the fore to accommodate design re-
`quirements. By any measure, good tools save valuable time.
`The more popular the architecture, the broader the array of
`tools written for use with it.
`Usually, these tools include an assembler, a simulator, a
`debugger, a compiler with a good library of routines, a link-
`er and librarian, and a library of common embedded-control
`routines. Although some of the tools come from the micro-
`controller vendors themselves, ample third-party support
`for the leading architectures gives customers an array of
`choices for implementing system designs. An entire indus-
`try of third-party tool vendors is flourishing around the larg-
`er and more widely used microcontroller architectures.
`
`…and on-chip memory
`More often than not, program memory is included on
`today’s embedded controller chips. These single-chip solu-
`
`The onward march of the microcontroller
`
`T he first true microcontrollers grew out of the Intel
`
`4004, a 4-bit microprocessor architecture devel-
`oped in the early 1970s. They ruled the industry
`until the late ‘80s, when 8-bit implementations dethroned
`them. Today, the microcontroller realm is still dominated
`by the 8-bit segment, which represents three-fifths of the
`total worldwide market.
`The most popular 8-bit architecture is the Motorola 68XX.
`Next comes the 80C51, originally developed by Intel but
`now offered by several vendors. In 1994, the top five sup-
`pliers of 8-bit microcontrollers were Motorola, NEC, Mitsub-
`ishi, Philips, and Hitachi.
`Designs employing 16-bit microcontrollers grew more
`common in the mid-to-late ‘80s, after the introduction of
`the architecture that still leads the 16-bit segment: the Intel
`80C196. Motorola (the 68H16), Philips (the XA), NEC (the K
`
`series) and other companies recently introduced new 16-bit
`architectures that are competing for market share.
`Intel’s i960 leads the way in the 32-bit market, which is
`as yet quite small but has been buoyed by a number of en-
`trants. These include Hitachi’s SH, Motorola’s Coldfire, and
`a slew of upcoming embedded reduced–instruction-set
`computer (RISC) offerings from such vendors as IDT, Tosh-
`iba, LSI Logic, Philips, and NEC, based on architectures from
`MIPS Technologies.
`As in all other segments of the microcontroller market-
`place, these new offerings give customers more choices,
`driving down prices and opening up new opportunities.
`Because the algorithms of digital signal processors (DSPs)
`are used in many embedded-control applications, DSPs are
`often regarded as a class of microcontroller, further expand-
`ing the range of today’s applications.
`—A.K.
`
`38
`
`IEEE SPECTRUM OCTOBER 1996
`
`Authorized licensed use limited to: Sterne Kessler Goldstein Fox. Downloaded on December 05,2022 at 20:39:11 UTC from IEEE Xplore. Restrictions apply.
`
`

`

`tions supply the need for small form factors and low cost,
`particularly in the 8-bit arena. If the task at hand warrants
`it, 32- and 16-bit controllers, which can address many
`megabytes of memory, may also use external memory.
`The least expensive type of memory available for this
`purpose is ROM. But, since ROM programming takes
`place during IC manufacture using a dedicated mask, any
`error means that a new mask must be generated. When
`that happens, design cycles drag on and costs start to bal-
`loon—not exactly to be desired in very cost- and time-
`sensitive business environments.
`To circumvent these difficulties, semiconductor suppliers
`often integrate more expensive one-time–programmable
`(OTP) ROM variants on chip: electrically programmable
`ROM and electrically erasable and programmable ROM.
`The first, EPROM, can be programmed electrically and
`erased by ultraviolet (UV) light. Putting these devices in
`plastic packages opaque to UV light produces OTP con-
`
`trollers that cannot be reprogrammed. EPROMs are cheap
`because they use very dense memory cells based on a single
`transistor. EEPROMs can be programmed and erased elec-
`trically but cost more because each cell requires the func-
`tional equivalent of an extra transistor.
`Often used in prototype development, reprogram-
`mable memories make it easier for designers to modify or
`upgrade code, providing a higher degree of flexibility.
`Their biggest advantage is a shorter time to market, ob-
`tained by eliminating the ROM masking cycle. Coupled
`with today’s sharply reduced product life cycles, this has
`popularized the use of OTP microcontrollers in embed-
`ded applications.
`Ideally, though, memory should be both reprogram-
`mable and inexpensive. While nothing today fully satisfies
`these dual requirements, most industry experts believe that
`flash memory may soon fit the bill. Flash mass storage is
`highly flexible and reprogrammable, and steady declines in
`
`[2] On average, the price
`of a microprocessor has
`more than tripled over the
`past five years, whereas a
`microcontroller sells for
`only 16 percent more. As
`a result, revenues from
`processor ICs have grown
`twice as fast as those from
`controller ICs even though
`unit sales of the controllers
`outpaced those of proces-
`sors by the same factor.
`The data shown here
`includes 4-bit microcon-
`trollers and all 8-, 16-, 32-,
`and 64-bit processors and
`controllers. 1996 figures are
`projected.
`
`16 000
`
`Microcontrollers
`Microprocessors
`
`12 000
`
`US $$
`
`Unitsales
`
`4
`
`3
`
`2
`
`1
`
`(x 106)
`
`Units
`
`(x 105)
`
`8000
`
`4000
`
`Revenues,millionsofU.S.dollars
`
`1991
`
`'92 '93 '94 '95 '96
`
`Source: Integrated Circuit Engineering Corp., 1995
`
`KHAN — WORKHORSES OF THE ELECTRONIC ERA
`
`39
`
`Authorized licensed use limited to: Sterne Kessler Goldstein Fox. Downloaded on December 05,2022 at 20:39:11 UTC from IEEE Xplore. Restrictions apply.
`
`

`

`associated costs indicate that the tech-
`nology will probably be cheaper to
`implement than EPROMs within the
`next three to five years. Today, many
`microcontroller vendors—including
`Philips, Motorola, Hitachi, Atmel, and
`Mitsubishi—are hard at work develop-
`ing flash-based products.
`
`12
`
`10
`
`Coming next
`Looking ahead, a number of tech-
`nologies are bound to influence the
`evolution of microcontrollers for em-
`bedded-control applications. As integ-
`ration marches on, submicrometer
`process technologies will affect system
`form factors and designs. Although to-
`day’s microprocessor processes are
`typically in the 0.35—0.5-␮m range,
`the cost sensitivities associated with
`microcontrollers currently dictate pro-
`cesses on the order of 0.8 ␮m. Within
`a few years, continuing advances pro-
`mise to drive microcontroller process
`technologies down to 0.5 ␮m and be-
`low. At process dimensions of this
`small size, providing and maintaining
`compatibility with nonvolatile technol-
`ogies will present manufacturers with a
`stiff challenge.
`Yet other pertinent technologies
`include the use of embedded dynamic
`RAM to provide for greater levels of
`system integration and more flexible
`options
`in system repartitioning;
`hardware description language (HDL)
`and synthesis techniques to save time
`and provide automated system simula-
`tions; and object-oriented program
`development to write code automati-
`cally in a way that is transparent to system product de-
`signers. Between them, all of these developments will
`make possible very large and complex single-chip con-
`trol systems, complete with memory and peripheral sub-
`systems, of a type that can be developed and program-
`med very quickly.
`Other hopeful trends are multichip packaging, micro-
`machining techniques that integrate sensors on chip, and
`the mapping of behavioral system descriptions into the
`microcontroller’s program. Ferroelectric memories in par-
`ticular bear watching. because they are nonvolatile and
`because, unlike dynamic RAMs, they do not need to be
`refreshed, while unlike electrically erasable and pro-
`grammable ROMs, they do not need special program-
`and-erase cycles. Unfortunately, the process of reading
`the cell is volatile, so information must be written back
`into memory. In addition, dedicated manufacturing lines
`are required to implement the technology. When these
`cells are able to read and write for around 1012 cycles,
`ferroelectric memories start to become an interesting
`alternative for embedded applications. Current ferroelec-
`tric technology is at about 108 cycles.
`These trends point to systems—all on a single chip—
`that will be able to sense such physical inputs as temper-
`
`8
`
`6
`
`Revenues,billions
`
`4
`
`2
`
`$8.2
`
`$6.6
`
`$5.2
`
`$4.85
`
`$11.7
`
`DSP
`
`16/32-
`bit
`
`$9.9
`
`8-bit
`
`4-bit
`
`1991 '92 '93 '94 '95 '96
`
`Source: Integrated Circuit Engineering Corp., 1995
`
`ature, acceleration, and light sensitivity; convert them
`into digital information; perform signal processing; out-
`put process-control signals; communicate status; handle
`remote diagnostics; reprogram themselves; and adaptive-
`◆
`ly change their functionality.
`
`To probe further
`The internals of a small real-time preemptive multitasking
`kernel are described in ␮C/OS, the Real-Time Kernel, by
`Jean J. Labrosse (R&D Publications, 1992, distributed by
`Prentice Hall).
`Journals and magazines that regularly cover microcontrollers
`include Electronic Engineering Times (CMP Publications,
`Manhasset, N.Y.), Microprocessor Report (MicroDesign Re-
`sources, Mountain View, Calif.), and EDN (Cahners Pub-
`lishing Co., Highlands Ranch, Colo.).
`
`About the author
`Ata R. Khan is manager of the Microcontroller Development
`Center at Philips Semiconductors, Sunnyvale, Calif. He is
`responsible for the strategic planning and architecture for
`microcontrollers.
`
`Spectrum editor: Linda Geppert
`
`[3] The 8-bit controller ICs bring in the lion’s share of total income from microcontrollers. But sales of
`16- and 32-bit microcontroller are growing rapidly. This year, digital signal processors are expected
`to bring in almost 14 percent of the total microcontroller income. The 1996 figures are projected.
`
`42
`
`IEEE SPECTRUM OCTOBER 1996
`
`Authorized licensed use limited to: Sterne Kessler Goldstein Fox. Downloaded on December 05,2022 at 20:39:11 UTC from IEEE Xplore. Restrictions apply.
`
`