BELL’S LAW FOR THEBIRTH AND DEATH OFCOMPUTER CLASSES

In the early 1950s, a person could walk inside a computer and by 2010 asingle computer (or “cluster’) with millions of processors will haveexpanded to the size of a building. More importantly, computers are begin-ning to “walk” inside of us. These ends of the computing spectrum illus-trate the vast dynamic range in computing power, size, cost, and otherfactors for early 21st century computer classes.

A computer class is a set of computers in a particular price range withunique or similar programming environments (such as Linux, OS/360,Palm, Symbian, Windows) that support a variety of applications that com-municate with people and/or other systems. A new computer class formsand approximately doubles each decade, establishing a new industry. Aclass may be the consequence and combination of a new platform with anew programming environment, a new network, and new interface withpeople and/or other information processing systems.

A theory of the computer’s evolution.

BY GORDON BELL

86 January 2008/Vol. 51, No. 1 COMMUNICATIONS OF THE ACM

COMMUNICATIONS OF THE ACM January 2008/Vol. 51, No. 1 87

Bell’s Law accounts for the formation, evolution, anddeath of computer classes based on logic technologyevolution beginning with the invention of the computerand the computer industry in the first-generation, vac-uum-tube computers (1950–1960), second-generation,transistor computers (1958–1970), through the inven-tion and evolution of the third-generation Transistor-Transistor Logic (TTL) and Emitter-coupled Logic(ECL) bipolar integrated circuits (ICs) from1965–1985. The fourth-generation MOS and CMOSICs enabling the microprocessor (1971) represents a“break point” in the theory because it eliminated theother early, more slowly evolving technologies. Moore’sLaw [6] is an observation about integrated circuit semi-conductor process improvements or evolution since thefirst IC chips, and in 2007 Moore extended the predic-

tion for 10–15 more years, as expressed in Equation 1.The evolutionary characteristics of disks, networks,displays, user interface technologies, and program-ming environments will not be discussed here.However, for classes to form and evolve, all tech-nologies must evolve in scale, size, and performanceat their own—but comparable—rates [5].

In the first period, the mainframe, followed by min-imal computers, smaller mainframes, supercomputers,and minicomputers established themselves as classes inthe first and second generations and evolved with thethird-generation integrated circuits circa 1965–1990.In the second or current period, with the fourth gen-eration, marked by the single processor-on-a-chip,evolving large-scale integrated circuits (1971–present)CMOS became the single, determinant technologyfor establishing all computer classes. By 2010, scalableCMOS microprocessors combined into powerful,multiple processor clusters of up to one million inde-pendent computing streams are likely. Beginning inthe mid-1980s, scalable systems have eliminated andreplaced the previously established, more slowly evolv-ing classes of the first period that used interconnectedbipolar and ECL ICs. Simultaneously smaller, CMOSsystem-on-a-chip computer evolution has enabledlow-cost, small form factor (SFF) or cell-phone-sizeddevices (CFSD); PDA, cell phone, personal audio(and video) devices (PADs), GPS, and camera conver-gence into a single platform will become the world-wide personal computer, circa 2010. Dust-sized chipswith relatively small numbers of transistors enable thecreation of ubiquitous, radio networked, implantable,sensing platforms to be part of everything and every-

one as a wireless sensor network class. Field Program-mable Logic Array chips with tens to hundreds of mil-lions of cells exist as truly universal devices for buildingnearly anything.

BELL’S LAW

A computer class is a set of computers in a particularprice range defined by: a programming environmentsuch as Linux or Windows to support a variety ofapplications; a network; and user interface for com-munication with other information processing sys-tems and people. A class establishes a horizontallystructured industry composed of hardware compo-nents through operating systems, languages, applica-tion programs and unique content includingdatabases, games, images, songs, and videos that serves

a market through various distribu-tion channels.

The universal nature of stored-program computers is such that acomputer may be programmed to

replicate function from another class. Hence, overtime, one class may subsume or kill off another class.Computers are generally created for one or more basicinformation processing functions—storage, computa-tion, communication, or control. Market demand fora class and among all classes is fairly elastic. In 2010,the number of units sold in classes will vary from tensfor computers costing around $100 million to billionsfor SFF devices such as cell phones selling for under$100. Costs will decline by increasing volume throughmanufacturing learning curves (doubling the totalnumber of units produced results in cost reduction of10%–15%). Finally, computing resources includingprocessing, memory, and network are fungible and canbe traded off at various levels of a computing hierarchy(for example, data can be held personally or providedglobally and held on the Web).

The class creation, evolution, and dissolutionprocess can be seen in the three design styles and pricetrajectories and one resulting performance trajectorythat threatens higher-priced classes: an establishedclass tends to be re-implemented to maintain its price,providing increasing performance; minis or minimal-cost computer designs are created by using the tech-nology improvements to create smaller computersused in more special ways; supercomputer design (thelargest computers at a given time come into existenceby competing and pushing technology to the limit tomeet the unending demand for capability); and theinherent increases in performance at every class,including constant price, threaten and often subsumehigher-priced classes.

All of the classes taken together that form the com-

88 January 2008/Vol. 51, No. 1 COMMUNICATIONS OF THE ACM

January Bell equation (1/08)

Transistors per chip = 2(t-1959) for 1959 ≤ t ≤ 1975; 216 x 2(t-1975)/1.5 for t ≥ 1975.

puter and communications industry shown in Figure 1behave generally as follows:

• Computers are born—classes come into existencethrough intense, competitive, entrepreneurial actionover a period of two to three years to occupy a pricerange, through the confluence of new hardware,programming environments, networks, interfaces,applications, and distribution channels. During theformation period, two to hundreds of companiescompete to establish a market position. After thisformative and rapid growth period, two or three, ora dozen primary companies remain as a class reachesmaturity depending on the class volume.

• A computer class, determined by a unique pricerange, evolves in functionality and graduallyexpanding price range of 10 maintains a stablemarket. This is followed bya similar lower-priced sub-class that expands therange another factor of fiveto 10. Evolution is similarto Newton’s First Law(bodies maintain theirmotion and directionunless acted on externally).For example, the “main-frame” class was establishedin the early 1950s using vacuum tube technologyby Univac and IBM and functionally bifurcatedinto commercial and scientific applications. Con-stant price evolution follows directly from Moore’sLaw whereby a given collection of chips providemore transistors and hence more performance.

A lower entry price, similar characteristics sub-class often follows to increase the class’s price rangeby another factor of five to 10, attracting more usageand extending the market. For example, smaller“mainframes” existed within five years after the firstlarger computers as sub-classes.

• CMOS semiconductor density and packaginginherently enable performance increase to supporta trajectory of increasing price and function.

Moore’s Law single-chip evolution, or micro-processor computer evolution after 1971 enablednew, higher performing and more expensive classes.The initial introduction of the microprocessor at asubstantially lower cost accounted for formation ofthe initial microcomputer that was programmed tobe a calculator. This was followed by more power-ful, more expensive classes forming including thehome computer, PC, workstation, the sharedmicrocomputer, and eventually every higher class.Home and personal computers are differentiated

from workstations simply on “buyer”—a personversus an organization.

The supercomputer class circa 1960 was estab-lished as the highest performance computer of theday. However, since the mid-1990s supercomput-ers are created by combining the largest number ofhigh-performance microprocessor-based computersto form a single, clustered computer system in asingle facility. In 2010, over a million processorswill likely constitute a cluster. Geographically cou-pled computers including GRID computing, suchas SETI@home, are outside the scope.

• Approximately every decade a new computer classforms as a new “minimal” computer either throughusing fewer components or use of a small fractionalpart of the state-of-the-art chips. For example, thehundredfold increase in component density per

decade enablessmaller chips, disks,and screens at thesame functionality ofthe previous decadeespecially since pow-erful microprocessorcores (for example,the ARM) use only afew (less than100,000) transistorsversus over a billionfor the largest Ita-nium derivatives.

Building the smallest possible computeraccounts for the creation of computers that wereused by one person at a time and were forerunnersof the workstation (for example, the Bendix G-15and LGP 30 in 1955), but the first truly personalcomputer was the 1962 Laboratory InstrumentComputer (LINC). LINC was a self-containedcomputer for an individual’s sole use with appropri-ate interfacial hardware (keyboards, displays), pro-gram/data filing system, with interactive programcreation and execution software. Digital Equipment’sPDP-1 circa 1961, followed by the more “minimal”PDP-5 and PDP-8 established the minicomputerclass [1] that was predominantly designed forembedded applications.

Systems-on-a-chip (SOCs) use a fraction of achip for the microprocessor(s) portion or “cores” tocreate classes and are the basis of fixed-functiondevices and appliances beginning in the mid-1990s.These include cameras, cell phones, PDAs, PADs,and their convergence into a single CPSD or SFFpackage. This accounts for the PC’s rapidly evolv-ing microprocessor’s ability to directly subsume the

COMMUNICATIONS OF THE ACM January 2008/Vol. 51, No. 1 89

Bell fig 1 (1/08)

Figure 1. Evolving computer classes based on technology anddesign styles: 1. constant price, increasing performance;

2. sub-class, lower price and performance to extend range; 3. supercomputer– largest computers that can be built that

extend performance; and 4. new, minimal, order ofmagnitude low-priced class formations every decade.

p

r

i

c

Time

3. Supercomputer: “the largest computers of the day”

4. New, “minimal priced” computers:smallest, useful computer, newapplications, new industry

1. Constant price, increasing performance. 2. Sub-class formation.

Figure 1. Evolving computer classesbased on technology and designstyles.

1980s workstation classby 1990.

• Computer classes die orare overtaken by lower-priced, more rapidlyevolving general-pur-pose computers as theless-expensive alterna-tives operating alone,combined into multipleshared memory micro-processors, and multi-ple computer clusters.Lower-priced platformsresult in more use andsubstantially higher vol-ume manufacturethereby decreasing costwhile simultaneouslyincreasing performancemore rapidly thanhigher-priced classes.

Computers can becombined to form a sin-gle, shared-memory com-puter. A “multi” ormultiple CMOS micro-processor, shared-memorycomputer [2] displacedbipolar minicomputerscirca 1990 and main-frames circa 1995, andformed the basic compo-nent for supercomputers.

Scalable, multiplecomputers can be net-worked into arbitrarilylarge computers to formclusters that replace cus-tom ECL and CMOSvector supercomputersbeginning in the mid-1990s simply becausearbitrarily large comput-ers can be created. Clus-ters of multiprocessors were called constellations;clusters using low latency and proprietary networksare MPPs (massively parallel processors).

Generality always wins. A computer created fora particular, specialized function, such as word pro-cessing or interpreting a language, and used for aparticular application is almost certain to be takenover by a faster-evolving, general-purpose com-puter. The computer’s universality property allows

any computer to takeon the function ofanother, given sufficientmemory and interfaces.

SFF devices sub-sume personal comput-ing functionality as theytake on the communi-cations functions of thePC (email and Webbrowsing), given suffi-cient memory andinterfaces. SFF devices,TVs, or kiosks accessingsupercomputers withlarge stores, subsumepersonal computingfunctionality. The largecentral stores retain per-sonal information, pho-tos, music, and video.

The specific characteristicsof the classes account forthe birth, growth, diminu-tion, and demise of variousparts of the computer andcommunications industry.

OVERVIEW OF THE BIRTH

AND DEATH OF THE

COMPUTER CLASSES

1951–2010The named classes andtheir price range circa2010 are given in Figure

2a. In 1986, David Nelson, the founder of Apollocomputer, and I posited that the price of a computerwas approximately $200 per pound [7]. Figure 2bgives the introduction price and date of the first ordefining computer of a class.

Here, I will use the aspects of Bell’s Law describedpreviously and follow a timeline of the class forma-tions beginning with the establishment of the firstcomputer classes (mainframe, supercomputer, sharedpersonal professional computers or workstations, andminicomputers) using vacuum tubes, transistors, andbipolar integrated circuits that continue through themid-1990s in the first period (1951–1990). In the sec-ond period beginning in 1971, the MOS micro-processor ultimately overtook bipolar by 1990 to

90 January 2008/Vol. 51, No. 1 COMMUNICATIONS OF THE ACM

Bell fig 2a (1/08)

Figure 2a. Computer classes and their price range 2005.

1.E+09

1.E+08

1.E+07

1.E+06

1.E+05

1.E+04

1.E+03

1.E+02

1.E+01

1.E+00

Con

sum

eran

dH

ome

Plat

form

sW

alle

tC

acul

ator

Pock

etC

acul

ator

Han

dhel

dga

me

(198

9)H

ome

Gam

eCon

sole

s(1

994)

Hom

ePC

(197

7)Pe

rson

alPl

atfo

rms

Prog

.Des

k/Po

cket

Cal

c,(1

9xx)

PDA

(199

5)Po

rtab

le,L

apto

pPC

(198

1)Pr

ofes

sion

alW

SSh

ared

,Dep

t.Sy

stem

sM

icro

com

pute

r(1

982)

Min

icom

pute

r(1

960)

Supe

rMin

icom

pute

rM

ini-m

ainf

ram

eC

entr

aliz

ed,S

yste

ms

Mai

nfra

me

Min

isup

erco

mpu

ter

Supe

rcom

pute

rSc

alab

leC

ompu

ting

Mul

tis(1

985)

Scal

able

s

Bell fig 2b (1/08)

Figure 2b. Introduction price versus date of the first orearly platforms to establish a computer class or lower priced

sub-class orginating from the same company or industry.

100,000

10,000

1,000

100

10

1

0.1

0.01

Pric

ein

1000

sof

$s

1940 1950 1960 1970 1980

Year of introduction of the first, earliest or dominant computer forming a class

1990 2000 2010

Figure 2a. Computer classesand their price range circa2005.

Figure 2b. Introduction price versus date of the first or early

platforms to establish a com-puter class or lower-priced

sub-class originating from thesame company or industry.

establish a single line based on CMOS technology.The section is followed by the three direct and indirecteffects of Moore’s Law to determine classes:

• Microprocessor transistor/chip evolution circa1971–1985 establish: calculators, home computers,personal computers, and workstations, and lower(than minicomputer) priced computers.

• “Minimal” designs establish new classes circa 1990that use a “fraction” ofthe Moore number.Microsystems evolutionusing fractional Moore’sLaw-sized SOCs enablesmall, lower-performing,minimal PC and commu-nication systems includ-ing PDAs, PADs,cameras, and cell phones.

• Rapidly evolving micro-processors using CMOSand a simpler RISCarchitecture appear as the“killer micro” circa 1985to have the same perfor-mance as supercomput-ers, mainframes,mini-supercomputers,super-minicomputers,and minicomputers builtfrom slowly evolving,low-density, customECL and bipolar inte-grated circuits. ECL sur-vived in supercomputersthe longest because of itsspeed and ability todrive the long transmis-sion lines, inherent inlarge systems. In theend, CMOS density andfaster system clocks overtook ECL by 1990.

The “killer micro” enabled by fast floating-pointarithmetic subsumed workstations and minicomputersespecially when combined to form the “multi” or mul-tiple microprocessor shared memory computer circa1985. “Multis” became the component for scalable clus-ters when interconnected by high-speed, low-latencynetworks. Clusters allow arbitrarily large computers thatare limited only by customer budgets. Thus scalabilityallows every computer structure from a few thousanddollars to several hundred million dollars to be arrangedinto clusters built from the same components.

In the same fashion that killer micros subsumed allthe computer classes by combining, it can be specu-lated that much higher volume—on the order of hun-dreds of millions—of SFF devices, may evolve morerapidly to subsume a large percentage of personal com-puting. Finally, tens of billions of dust-sized, embedd-able wirelessly connected platforms that connecteverything are likely to be the largest class of allenabling the state of everything to be sensed, effected,

and communicated with.

MICROPROCESSORS CIRCA

1971: THE EVOLVING

FORCE FOR CLASSES IN

THE SECOND PERIOD

Figure 3 shows the micro-processors derived directlyfrom the growth of tran-sistors/chips beginning in1971. It shows the trajec-tory of microprocessorsfrom a 4-bit data paththrough, 8-, 16-, 32-, and64-bit data paths andaddress sizes. The figureshows a second path—theestablishment of “mini-mal” computers that useless than 50 thousandtransistors for the proces-sor, leaving the remainder

of the chip for memory and other functions (forexample, radio, sensors, analog I/O) enabling thecomplete SOC. Increased performance, not shown inthe figure, is a third aspect of Moore’s Law that allowsthe “killer micro” formation to subsume all the other,high-performance classes that used more slowlyevolving bipolar TTL and ECL ICs. Calculators,home computers, personal computers, and worksta-tions were established as classes as the processor on achip evolved to have more transistors with wide datapaths and large address spaces as shown in Figure 3.

In 1971, Intel’s 4004 with a 4-bit data path andability to address 4KB was developed and pro-grammed to be the Busicom Calculator; instead ofdeveloping a special chip as had been customary toimplement calculators, a program was written for the4004 for it to “behave” as or “emulate” a calculator.

In 1972, Intel introduced the 8008 microprocessorcoming from the Datapoint terminal requirement,with a 8-bit data path and ability to access 16KB thatallowed limited, programmable computers followed bymore powerful 8080-based systems MITS used tointroduce its Altair personal computer kit in 1975,

COMMUNICATIONS OF THE ACM January 2008/Vol. 51, No. 1 91

Bell fig 3 (1/08)

Figure 3. Moore’s Law that provides more transistors per chip,has resulted in creating the following computer classes:

calculators, home computers, personal computers,workstations, “multis” to overtake minicomputers,

and clusters using multiple core, multi-threading to overtakemainframes and supercomputers.

10

100

1,000

10,000

100,000

Intel Itanium

MulticoreMultithreadSPARC

64-bit Alpha

Intel 32-bit 486386

68000 Apollo Sun Workstation

One Million

Moore’sLaw

8088 IBM PCARM Embedded and other

System-on-a-chip computers6502 Apple, Commodore

8080 Altair8008 Micral, Scelbi

4004: Busicom Calculator

1,000,000

10,000,000

Tran

sist

ors

(100

0s)

ofea

chm

icro

proc

esso

ror

mic

roco

mpu

ter

4004: BusicomCalculator

0

1960 1970 1980 1990 2000 2010

Figure 3. Moore’s Law, whichprovides more transistors perchip per year, has resulted in

creating the following computerclasses: calculators, home

computers, personal computers,workstations, “multis” to

overtake minicomputers, andclusters using multiple core,

multithreading to overtake mainframes and

supercomputers.

which incidentally stimulated Gates and Allen to startMicrosoft. In 1977, the 16-bit 6502 microprocessorand higher-capacity memory chips enabled personalcomputers for use in the home or classroom built byApple, Commodore, and Radio Shack—computersthat sold in the tens of millions because people boughtthem to use at home versus corporate buyers. By 1979,the VisiCalc spreadsheet ran on the Apple II establish-ing it as a “killer app” for personal computers in awork environment. Thus, the trajectory went from a4-bit data path and limited address space to a 16-bitdata path with the ability to access 64KB of memory.This also demonstrates the importance of physicaladdress as an architectural limit. In the paper onDEC’s VAX [3], we described the importance ofaddress size on architecture: “There is only one mis-take that can be made in a computer design that is dif-ficult to recover from—not providing enough addressbits for memory addressing and memory manage-ment…” The 8086/8088 of the first IBM PCs had a20-bit, or 1MB address space, the operating systemusing the remaining 384KB.

Concurrent with the introduction of the IBM PC,professional workstations were being created thatused the Motorola 68000 CPU with its 32-bit dataand address paths (4GB of maximum possible mem-ory). Apple Computer used the Motorola “68K” inits Lisa and Macintosh machines. IBM’s decision touse the Intel architecture with limited addressing,undoubtedly had the effect of impeding the PC by adecade as the industry waited for Intel to evolvearchitecture to support a larger address and virtualmemory space. Hundreds of companies started up tobuild personal computers (“PC clones”) based on theIBM PC reference design circa 1981. Dozens of

companies also started to build workstations basedon a 68K CPU running the UNIX operating system.This was the era of “JAWS” (Just Another WorkSta-tion) to describe efforts at Apollo, HP, IBM, SGI,Sun Microsystems and others based on 32-bit versus16-bit. Virtually all of these “workstations” wereeliminated by simple economics as the PC—basedon massive economies of scale and commoditizationof both the operating system and all constituenthardware elements—evolved to have sufficient powerand pixels.

“Minimal” CMOS Microsystems on a Chip circa1990 Establish New Classes using Smaller, Less-Expensive, Chips. In 2007, many systems are com-posed of microprocessor components or “cores” withless than 50,000 transistors per microprocessor core ata time when the leading-edge microprocessor chipshave a billion or more transistors (see Figure 3). Suchcores using lower cost, less than the state-of-the-artchips and highly effective, rapid design tools allow new,minimal classes to emerge. PDAs, cameras, cell phones,and PADs have all been established using this minimalcomputer design style based on small cores. In 1990,the Advanced RISC Machine (ARM) formed from acollaboration between Acorn and Apple as the basis forembedded systems that are used as computing plat-forms and achieved two billion units per year in 2006.Other higher-volume microsystem platforms using 4-,8-…64-bit architectures including MIPS exist as corearchitectures for building such systems as part of thevery large embedded market.

Rapidly Evolving Killer CMOS Micros circa 1985Overtake Bipolar ICs to Eliminate EstablishedClasses. In the early 1980s, the phrase “killer micro”was introduced by members of the technical comput-

92 January 2008/Vol. 51, No. 1 COMMUNICATIONS OF THE ACM

Over time, the performance of lesser-performing, faster-evolving products eventually overtakes the established, slowly evolving classes served by sustaining technology.

ing community as they saw how the more rapidlyevolving CMOS micro would overtake bipolar-basedminicomputers, mainframes, and supercomputers ifthey could be harnessed to operate as a single systemand operate on a single program or workload.

In the Innovator’s Dilemma, Christensen describesthe death aspect basis of Bell’s Law by contrasting twokinds of technologies [4]. Sustaining technology pro-vides increasing performance, enabling improvedproducts at the same price as previous models usingslowly evolving technology; disruptive, rapidly evolv-ing technology provides lower-priced products that arenon-competitive with higher-priced sustaining class tocreate a unique market space. Over time, the perfor-mance of lesser-performing, faster-evolving productseventually overtakes the established, slowly evolvingclasses served by sustaining technology.

From the mid-1980s until 2000, over 40 companieswere established and went out of business attempting toexploit the rapidly evolving CMOS microprocessors byinterconnecting them in various ways. Cray, HP, IBM,SGI, and Sun Microsystems remain in 2008 to exploitmassive parallelism through running a single programon a large number of computing nodes.

Two potentially disruptive technologies for newclasses include:

• The evolving SFF devices such as cell phones arelikely to have the greatest impact on personal com-puting, effectively creating a class. For perhaps mostof the four billion non-PC users, a SFF devicebecomes their personal computer and communica-tor, wallet, or map, since the most common andoften only use of PCs is for email and Web brows-ing—both stateless applications.

• The One Laptop Per Child project aimed at a $100PC (actual cost $188 circa November 2007) is pos-sibly disruptive as a “minimal” PC platform withjust a factor-of-two cost reduction. This is achievedby substituting 1G of flash memory for rotating-disk-based storage, having a reduced screen size, asmall main memory, and built-in mesh networkingto reduce infrastructure cost, relying on the Internetfor storage. An initial selling price of $188 for theOLPC XO-1 model—approximately half the priceof the least-expensive PCs in 2008—is characteristicof a new sub-class. OLPC will be an interestingdevelopment since Microsoft’s Vista requires almostan order of magnitude more system resources.

FUTURE CHALLENGES

The Challenge of Constant Price, 10–100 billionTransistors per Chip, for General-Purpose Com-puting. The future is not at all clear how such large,

leading-edge chips will be used in general-purposecomputers. The resilient and creative supercomput-ing and large-scale service center communities willexploit the largest multiple-core, multithreadedchips. There seems to be no upper bound these sys-tems can utilize. However, without high-volumemanufacturing, the virtuous cycle is stopped—inorder to get the cost and benefit for clusters, a high-volume personal computer market must drivedemand to reduce cost. In 2007, the degree of paral-lelism for personal computing in current desktop sys-tems such as Linux and Vista is nil, which eitherindicates the impossibility of the task or the inade-quacy of our creativity.

Several approaches for very large transistor count(approximately 10 billion transistor chips) could be:

• System with primary memory on a chip forreduced substantially lower-priced systems andgreater demands;

• Graphics processing, currently handled by special-ized chips, is perhaps the only well-defined appli-cation that is clearly able to exploit or absorbunlimited parallelism in a scalable fashion for themost expensive PCs (such as for gaming andgraphical design);

• Multiple-core and multithreaded processor evolu-tion for large systems;

• FPGAs that are programmed using inherently par-allel hardware design languages like parallel C orVerilog that could provide universality that wehave not previously seen; and

• Interconnected computers treated as softwareobjects, requiring new application architectures.

Independent of how the chips are programmed,the biggest question is whether the high-volume PCmarket can exploit anything other than the first pathin the preceding list. Consider the Carver Mead 11-year rule—the time from discovery and demonstra-tion until use. Perhaps the introduction of a fewtransactional memory systems has started the clockusing a programming methodology that claims to bemore easily understood. A simpler methodology thatcan yield reliable designs by more programmers isessential in order to utilize these multiprocessor chips.

Will SFF Devices Impact Personal Computing?Users are likely to switch classes when the perfor-mance and functionality of a lesser-priced class is ableto satisfy their needs and still increase functionality.Since the majority of PC use is for communicationand Web access, evolving a SFF device as a singlecommunicator for voice, email, and Web access isquite natural. Two things will happen to acceleratethe development of the class: people who have never

COMMUNICATIONS OF THE ACM January 2008/Vol. 51, No. 1 93

used or are without PCs will use the smaller, simplerdevices and avoid the PC’s complexity; and existingPC users will adopt them for simplicity, mobility, andfunctionality. We clearly see these small personaldevices with annual volumes of several hundred mil-lion units becoming the single universal device evolv-ing from the phone, PDA, camera, personalaudio/video device, Web browser, GPS and map, wal-let, personal identification, and surrogate memory.

With every TV becoming a computer display, acoupled SFF becomes the personal computer for theremaining applications requiring large screens. Cablecompanies will also provide access via this channel asTV is delivered digitally.

Ubiquitous Wireless: WiFi, Cellular Services, andWireless Sensor Nets. Unwiring the connectionaround the computer and peripherals, televisions, andother devices by high-speed radio links is useful butthe function is “unwiring,” and not platform creation.Near-Field Communication (NFC) using RF or mag-netic coupling offers a new interface that can be usedto communicate a person’s identity that could form anew class for wallets and identity. However, mostlikely the communication channel and biometric tech-nology taken together just increase the functionality ofsmall devices.

Wireless Sensor Nets: New Platform, Network,and Applications. Combining the platform, wirelessnetwork, and interface into one to integrate withother systems by sensing and effecting is clearly anew class that has been forming since 2002 with anumber of new companies that are offeringunwiring, and hence reduced cost for existing appli-cations, such as process, building, home automation,and control. Standards surrounding the 802.15.4link that competes in the existing unlicensed RFbands with 802.11xyz, Bluetooth, and phone trans-mission are being established.

New applications will be needed for wireless sensornets to become a true class versus just unwiring theworld. If, for example, these chips become part ofeverything that needs to communicate in the whole IThierarchy, a class will be established. They carry outthree functions when part of a fixed environment or amoving object: sense/effect; recording of the state of aperson or object (things such as scales, appliances,switches, thermometers, and thermostats) includingits location and physical characteristics; and commu-nication to the WiFi or other special infrastructurenetwork for reporting. RFID is part of this potentiallyvery large class of trillions. Just as billions of clientsneeded millions of servers, a trillion dust-sized wirelesssensing devices will be coupled to a billion other computers.

CONCLUSION

Bell’s Law explains the history of the computingindustry based on the properties of computer classesand their determinants. This article has posited ageneral theory for the creation, evolution, and deathof various priced-based computer classes that havecome about through circuit and semiconductortechnology evolution from 1951. The exponentialtransistor density increases forecast by Moore’s Law[6] being the principle basis for the rise, dominance,and death of computer classes after the 1971 micro-processor introduction. Classes evolve along threepaths: constant price and increasing performance ofan established class; supercomputers—a race tobuild the largest computer of the day; and novel,lower-priced “minimal computers.” A class can besubsumed by a more rapidly evolving, powerful, less-expensive class given an interface and functionality.In 2010, the powerful microprocessor will be thebasis for nearly all classes from personal computersand servers costing a few thousand dollars to scalableservers costing a few hundred million dollars. Com-ing rapidly are billions of cell phones for personalcomputing and the tens of billions of wireless sensornetworks to unwire and interconnect everything. AsI stated at the outset, in the 1950s a person couldwalk inside a computer and by 2010 a computercluster with millions of processors will haveexpanded to the size of a building. Perhaps more sig-nificantly, computers are beginning to “walk” insideof us.

References1. Bell, C.G. The mini and micro industries. Computer 17, 10 (Oct. 1984),

14–30.2. Bell, C.G. Multis: A new class of multiprocessor computers. Science 228

(Apr. 26, 1985) 462–467. 3. Bell, G. and Strecker, W. Computer structures: What have we learned

from the PDP-11. IEEE Computer Conference Proceedings (Florida, Nov.1975).

4. Christensen, C.M. The Innovator’s Dilemma. Harvard Business SchoolPress, 1997.

5. Gray, J. and Shenoy, P. Rules of thumb in data engineering. In Pro-ceedings of ICDE200 (San Diego, Mar. 1–4, 2000). IEEE press.

6. Moore, G.E. Cramming more components onto integrated circuits.Electronics 8, 39 (Apr. 19, 1965); revised 1975.

7. Nelson, D.L. and Bell, C.G. The evolution of workstations. IEEE Cir-cuits and Devices Magazine (July 1986), 12–15.

Gordon Bell (gbell@microsoft.com) is a principal researcher inMicrosoft Research Silicon Valley, working in the San Francisco Laboratory.

Permission to make digital or hard copies of all or part of this work for personal or class-room use is granted without fee provided that copies are not made or distributed forprofit or commercial advantage and that copies bear this notice and the full citation onthe first page. To copy otherwise, to republish, to post on servers or to redistribute tolists, requires prior specific permission and/or a fee.

© 2008 ACM 0001-0782/08/0100 $5.00

c

94 January 2008/Vol. 51, No. 1 COMMUNICATIONS OF THE ACM