A microprocessor (from the Greek μικρός mikrós , German 'small, tight' ) is a very small-scale processor in which all the components of the processor are combined on a microchip ( integrated circuit , IC). The first microprocessor was developed in the early 1970s by Texas Instruments on the basis of IC technology.
|Number of transistors
|Intel Pentium M
(with 9 MB cache )
|Sony / IBM / Toshiba
|Core 2 Duo
|Core 2 Quad
|Dual-core Itanium 2
|Intel Core i7 2600K
|Intel Core i7 3930K
|Intel Core i7 4770K
In the early 1960s, tube processors were replaced by transistorized types. Initially, the processors were built discreetly from individual tubes. Such a processor had the volume of a wall cabinet , the power consumption was a few thousand watts , the clock frequency 100 kHz.
The technological leap from tube to transistor technology resulted in a lower space requirement, lower temperature development, higher processing speed, a lower failure rate and lower power consumption of only a few 100 watts. The clock frequency increased to about 1 MHz. The later reduction in size of the transistors to just a few micrometers made it possible to accommodate more and more transistor functions on integrated circuits (ICs). At first it was just individual gates , but more and more complete registers and functional units such as adders and counters, and finally even register banks and arithmetic units, were integrated on a chip. This increasing integration of more and more transistor and gate functions on one chip then almost inevitably led to what is now known as a microprocessor.
The microprocessor was patented by employees of Texas Instruments (TI), who presented the circuit called TMS1000 in 1971 . In addition to a main processor, it contained a 1 KiB ROM , a 64 × 4-bit RAM and other functions such as counters and timers as well as interfaces for input and output . At that time, these were usually implemented in individual circuits and the TMS1000 therefore corresponds to a microcontroller . In the same year as TI, Intel presented the "microprocessor" (English microprocessor unit , MPU) with the 4004 , which is regarded as the first main processor (CPU) on a chip, since TI did not market the TMS1000 until 1974 as an independent product. With only 4- bit wide registers and a clock frequency of up to 740 kHz, the 4004 was not very powerful. However, its extremely compact design compared with the classic CPUs ultimately helped the microprocessor achieve its breakthrough. Originally the 4004 was an order development for the Japanese desk calculator manufacturer Busicom . In 1969, Intel's Ted Hoff , head of the Application Research department , had the idea of realizing the heart of this desktop computer in the form of a programmable module. In 1970 Federico Faggin , in the department for investigations into the metal-insulator-semiconductor structure , developed a circuit integration based on transistors with a gate electrode made of silicon for the implementation of the 4004 and led the project to its successful debut in the market Year 1971. It was not actually intended that this would result in the world's first universally applicable single-chip CPU. Since Busicom was in financial difficulties at the time, they offered Intel to buy back the 4004 design, whereupon Intel began marketing the 4004. The 4004 became the world's first commercial microprocessor. Another notable development was only discovered in 1998 after military records were released. According to this, Garrett AiResearch (among others with employees Steve Geller and Ray Holt ) developed a chipset (system consisting of several ICs, including CPU) for military purposes between 1968 and 1970 . The chipset known as MP944 was part of the Central Air Data Computer (CADC), the flight control system of the new F-14 Tomcat ( US Navy ).
At first these were quite simple circuits . The microelectronics brought other advantages such as speed, low power consumption, reliability, and later more complex next to the miniaturization and cost saving. As a result, comparatively cheap microprocessors over time replaced the expensive processors in minicomputers and in some cases even in mainframes . Towards the end of the 20th century, the microprocessor found its way into many electronic devices, especially as the CPU of personal computers (PCs). Even when the structural size of the microprocessor chips was further reduced to a few nanometers (14 nm, as of January 2015, Intel Broadwell architecture ), the term microprocessor remained.
The word length was initially limited to 4 bits because of the not yet so high integration density. As a result, the word length was continuously increased, mostly in doubling steps. Since the resources were still so expensive at the beginning, they looked for ways to optimally adapt them to the respective requirements. One episode along the way were bit-slice systems, in which several bit-slice processors with a small bit width could be interconnected to form a system with the desired, larger bit width.
To implement a complete computer , the microprocessor still has to be expanded to include memory and input / output functions . These are available in the form of additional chips. Only a few years after the introduction of microprocessors, the term microcontroller was established , which combines these functions on one chip.
Notable 8-bit processors
The 4004 was replaced in 1972 by the 8008 , the world's first 8-bit microprocessor. This processor was the forerunner of the extremely successful Intel 8080 (1974) and other 8-bit processors from Intel. The competing Motorola 6800 was available from August 1974, the same year as the 8080. The architecture of the 6800 was copied and improved in 1975 for the MOS Technology 6502 , which rivaled the Z80 in popularity in the 1980s.
The development team of the 8080 founded the company Zilog and brought out the Z80 in 1976, a greatly improved and code-compatible further development. This achieved the greatest popularity of all 8-bit processors. For details see Zilog Z80 .
Both the Z80 and 6502 were designed with a low total cost in mind. The housing was small, the demands on the bus were low and circuits were integrated that previously had to be provided in a separate chip (the Z80, for example, had its own refresh generator for dynamic RAM memory DRAM ). It was these features that finally helped the home computer market break through in the early 1980s, resulting in machines that went for $ 99.
The SC / MP was sold by the Santa Clara - based National Semiconductor Corporation in the mid-1970s. Various single-board computers were built as do-it-yourself and teaching computers based on the SC / MP until around 1980.
Western Design Center (WDC) introduced the CMOS 65C02 in 1982 and licensed the design to various companies. This processor became the heart of the Apple IIc and IIe and was used in pacemakers and defibrillators , automobiles, as well as industrial equipment and the consumer market. WDC pioneered the licensing of microprocessor technology; this business model was later adopted by ARM and other manufacturers in the 1990s.
Motorola trumped the entire 8-bit world in 1978 with the introduction of the Motorola 6809 , one of the most powerful and cleanest 8-bit architectures and also one of the most complex microprocessor logics ever produced. At that time, microprogramming replaced the previously hard-wired logics - precisely because the design requirements for hard wiring were becoming too complex.
A pioneering microprocessor for space travel was the RCA1802 (alias CDP1802, RCA COSMAC; introduced in 1976), which was used in the Voyager , Viking and Galileo space probes . The CDP1802 was used because it could be operated with very little energy and its construction ( Silicon-on-Sapphire ) offered a much higher protection against cosmic rays and electrostatic discharges than any other processor at the time. The CP1802 has been called the first radiation-hardened processor.
The first multi-chip 16-bit microprocessor was the IMP-16 from National Semiconductor, introduced in 1973. An 8-bit version followed a year later as the IMP-8 . In 1975 National Semiconductor introduced the first single-chip 16-bit microprocessor PACE , which was later followed by an NMOS version ( INS8900 ).
Other multi-chip 16-bit microprocessors included TI's TMS 9900, which was also compatible with the in-house TI-990 line of minicomputers . The chip was packaged in a large 64-pin DIP package , while most of the 8-bit processors were housed in the more popular, smaller, and cheaper 40-pin plastic DIP package. A successor was developed from the 9900, the TMS 9980, which also had a cheaper housing. It should be a competitor to the Intel 8080. The TMS 9980 could copy 8 data bits at the same time, but only address 16 KiB . A third chip, the TMS 9995, has been newly developed. This processor family was later expanded with the 99105 and 99110.
WDC made its 65C02 16-bit compatible and introduced this processor in 1984 as CMOS 65816. The 65816 represented the core of the Apple IIgs and later the Super Nintendo , making it one of the most popular 16-bit designs.
Intel adopted a strategy of not emulating minicomputers and instead “enlarged” its 8080 design to 16 bits. This resulted in the Intel 8086 , the first member of the x86 family that can be found in most PCs today. The 8086 and its "little brother", the 8088 , offered the option of porting software from the 8080 line, which Intel did good business with. The successor to the 8086 was the 80186 , the 80286 and, in 1985, the 32-bit 80386 processor , all of which were backwards compatible and thus decisively strengthened Intel's market dominance.
The first 32-bit microprocessor in its own case was the BELLMAC-32A from AT&T Bell Labs , the first pieces of which were available in 1980 and which were mass-produced in 1982. After AT&T was broken up in 1984, it was renamed WE 32000 (WE for Western Electric) and had two successors: the WE 32100 and WE 32200. These microprocessors were used in the following AT&T minicomputers: 3B2, 3B5, 3B15, "Companion" and "Alexander".
One of the bemerkenswertesten 32-bit microprocessors, the MC68000 from Motorola , which was presented 1979th Often referred to as the 68K, it had 32-bit registers, but used 16-bit internal bus lines and an external data bus that was equally wide to reduce the number of pins required . Motorola generally referred to this processor as a 16-bit processor, even though it had a 32-bit architecture internally. The combination of a fast and large memory address space (16 megabytes) and low cost made it the most popular processor in its class. The Apple Lisa and the Macintosh series used the 68K; In the mid-1980s this processor was also used in the Atari ST and Commodore Amiga .
Intel's first 32-bit microprocessor was the iAPX 432, introduced in 1981 . Although it had an advanced, object-oriented architecture, it did not achieve any commercial success - not least because it performed worse than competing architectures.
Motorola's success with the 68K led to the launch of the MC68010 , which supports virtual memory addressing technology . The MC68020 eventually had 32-bit internal and external buses. This processor became extremely popular in the Unix super microcomputer, and many smaller companies made desktop systems with this processor. The MC68030 integrated the MMU into the chip. Most computers that didn't run on DOS now use a 68K family chip. This continued success led to the MC68040 , which also integrated the FPU into the chip, increasing the speed of arithmetic operations . A planned MC68050 did not achieve the desired improvements and was not produced, the MC68060 was thrown onto a market segment that was already saturated with much faster RISC designs.
The 68020 and its successors were widely used in embedded systems .
During this time (early to mid-1980) National Semiconductor manufactured a 32-bit processor with a 16-bit pinout, similar to Motorola, the NS 16032 (later renamed NS 32016). The version with a 32-bit bus was the NS 32032. Sequent introduced the first SMP computer based on this microprocessor in the mid- 1980s .
Other systems used the Zilog Z80000 , but it arrived too late in the market and soon disappeared again.
64-bit processors on the desktop
While 64-bit processors had been in use in various markets since the early 1990s, they were only used on the PC market after 2000. In July 2003, Apple presented the Power Mac G5, Apple's first 64-bit desktop computer , at the Developer Conference (WWDC) . Before that, 64-bit computers were already available from Sun and other manufacturers, but these are usually referred to as workstations and not as desktop computers, even if no technical feature justifies this distinction.
Around the same time, with AMD's introduction of the first 64-bit architecture AMD64 ( backwards compatible with IA-32 ) in September 2003, the era of 64-bit architectures also began for x86 computers. AMD was soon followed by Intel, which had previously produced IA-64 CPUs ( Intel Itanium ), but which failed due to the lack of backward compatibility in the consumer market. Now Intel turned to the AMD64 architecture and has been producing its own AMD64-compatible x86 / Intel64 processors since the Intel Pentium 4 core Prescott 2M (Release: February 2005). Both x86 processors can run the previous 32-bit software as well as the new 64-bit software. With 64-bit Windows XP and Linux , the software is now moving towards the new architecture and using the full potential of these processors. Since the release of Windows 7 , most OEM computers have been released with a 64-bit version, especially since the magical 4 GB RAM limit of 32-bit systems in commercial computers was reached around 2010 .
Especially with the IA-32, the change to 64-bit is more than just increasing the register width, as the number of registers has also been increased.
In the case of PowerPC architectures, the switch to 64-bit was made in the early 1990s (in fact, the PPC processor is designed from the outset as 64-bit, with a 32-bit subset of commands). The register sizes and internal buses are increased, the arithmetic and vector arithmetic units worked with 64 or more bits for several years before the change (this is also the case with IA-32). However, no new registers are inserted, which means that the speed gained from 64 to 32-bit is lower than with IA-32.
In the mid-1980s to the early 1990s, many RISC microprocessors ( Reduced Instruction Set Computing ) appeared, which were initially used in specialized computers and UNIX workstations, but have since been used universally in a wide variety of tasks; Intel standard desktop computers today are mixed forms of RISC-CISC.
The first commercial architecture came from MIPS Technologies , the R2000 (the R1000 was not sold). The R3000 made the architecture really practical, the R4000 ultimately represented the world's first 64-bit architecture. Competing projects produced the IBM Power and Sun SPARC systems. Soon every major manufacturer had a RISC design on offer, e.g. B. the AT&T CRISP , AMD Am29000 , Intel i860 and Intel i960 , Motorola 88000 , DEC Alpha and the HP PA-RISC .
Competition soon made most of these architectures disappear, leaving IBM's POWER and its derived PowerPC (as the desktop RISC architecture) and Sun SPARC (only in Sun's own systems). MIPS continues to offer SGI systems, but the architecture is mostly used as an embedded design , e.g. B. in the routers from Cisco .
Other companies focus on niche markets, especially ARM Limited , which was spun off from Acorn in 1989 . Acorn was a manufacturer of RISC computers that was one of the first to target the home computer market with the ARM architecture- based model series Acorn Archimedes and Acorn Risc PC . ARM is now concentrating on processors (see also ARM architecture) for embedded systems.
Design and manufacture
A microprocessor is a processor in which all components of the processor are combined on a microchip . The microprocessors are adapted to the respective area of application due to their different areas of application. For example, special versions have air - and aerospace particularly high temperatures and radiation exposure to withstand error-free in operation, while mobile processors a high IPC -rate, low leakage currents must have and low power consumption. These needs are taken into account in various ways: A fundamental design decision is already made with the selection of the instruction set ( CISC or RISC ), the implications of which are explained in more detail in the respective special articles. A microcode that is as efficient as possible is then developed, which is optimally adapted to boundary conditions such as cache sizes, memory bandwidth and latencies as well as the internal functional units.
The logical design of the microprocessor, which is available in a hardware description language , is then transferred to a high-performance computer, which "routes" the conductor tracks , i. That is, seeks to determine an optimal arrangement with as few transistors as possible and minimal power loss (so-called technology binding or technology mapping ). Since these routing problems are NP-complete , only approximate solutions are usually found that can be considerably improved in detail. From these path calculations, masks are created that are used by means of photolithography to expose wafers , which are then etched. The production of today's microprocessor comprises well over 100 individual steps, in the course of which a single mistake can render the entire processor unusable.
In the final inspection, the processors are finally classified with regard to their cycle stability, with physical properties such as signal levels at different cycles being checked using a test program developed individually for each processor type. Here, especially in term of critical paths on the CPU The attention to speed Paths to prevent (errors caused by signal delays).
In general, it can be stated that the validation effort of modern processors has assumed enormous proportions, and despite all efforts, not all error situations can be checked before delivery. The last x86 processor fully verified in all functions (and errors!) Was the 80286 . For this reason, all manufacturers provide so-called errata lists in which discovered errors are recorded. For example, Intel had to admit the notorious FDIV bug in early Pentium CPUs, which can be traced back to several missing entries in an internal lookup table of the FPU .
In the course of time, the number of instructions supported by the processor increased due to the constantly improving technology. Today there are predominantly 32 and 64 bit processors, whereby the most common operating systems for the user support a maximum of 64, but mostly only 32 bit. This already shows that the software lags behind the hardware in the case of processors. The 386s developed in the 1980s were the first 32-bit processors in the Intel 80x86 family.
In 2006 the ARM company presented the first commercial non-clocked, asynchronous processor , the ARM996HS. Since it works without clocking, an asynchronous processor has a lower and significantly less pronounced radiation in the high frequency range and does not consume any significant current during process breaks.
In the course of ever increasing integration densities of semiconductor processes , the developers of CPUs have integrated additional functions into the hardware. The units that previously had to be connected as separate chips and that over time could be integrated into the CPU itself include:
- the memory management unit for memory management;
- the numerical coprocessor for faster arithmetic operations with floating point numbers ;
- Vector arithmetic units , especially for fast graphics processing - at Intel under the names MMX , SSE and successors, at PowerPC as AltiVec ;
- Cache memory, initially only level 1, today also level 2 and already level 3;
- sometimes the chipset (or parts of it) to control the main memory ;
- sometimes a graphics chip for display control;
- up to 100 processor cores on one chip ( multi-core processor , terascale processor ).
Microcontrollers, on the other hand, often have only a few registers and a limited set of instructions, where addition and subtraction are often the most complex operations. For simple applications, such as controlling a simple machine, this functionality is sufficient, especially since higher functions can be implemented with a few basic operations, for example multiplication by shifting and adding (see Russian pawn multiplication ). For this purpose, microcontrollers integrate peripheral functions and often also working memory on the chip.
In connection with rising electricity costs, the energy consumption of microprocessors is becoming an increasingly important performance characteristic. This applies above all to mainframes , data centers and server farms as well as to mobile devices such as smartphones or tablet computers . Energy-saving processors also offer advantages outside of data centers. Since the coolers have less to do, the computers are also quieter. Sometimes the computers can even be passively cooled. And in summer, the heat generated by a PC in a room without air conditioning is a nuisance for the people present there.
In the past, new manufacturing techniques (especially structural reductions ) were used primarily to increase frequency; In the meantime, they are partly used to reduce the previously steadily increasing power consumption:
- Instead of ever higher clock rates for a single computing core, several computing cores are now accommodated in one processor at lower frequencies.
- Optimized manufacturing reduces leakage currents .
Current multi-core processors can have a power requirement between 45 and 140 watts ( TDP ), depending on the model . Energy-saving capabilities are also increasingly being built in so that components that are not required can be clocked more slowly at times or switched off completely. Concerning. of the total power consumption is i. A. the race-to-idle principle applied. Modern processors sometimes even have a “ turbo mode ” in order to fully utilize the available cooling reserves.
The power consumption of processors is burdened with further follow-up costs: The electricity used is converted into heat, which has to be removed from the computer by the fan. Higher consumption requires more powerful fans, which also use more electricity. If the location of the computer itself is an air-conditioned room, the air-conditioning system is also burdened. Depending on the coefficient of performance of the cooling device , you can expect approx. 25–40% additional consumption , i. H. a 300 W PC loads the air conditioning system with 75–120 W more power. The computer's power supply unit may also have to be larger. If the computer is connected to a UPS , this also has a higher internal consumption, depending on its efficiency. With many computers in one place, additional investment costs for larger air conditioning systems and larger UPS systems can arise. Servers usually run 24 hours a day, seven days a week, for a total of 8760 hours a year. In order to improve the energy balance of IT systems, different approaches are being pursued. The aim is to increase the effectiveness of the cooling (example: Air Guide ) and to use the heat given off (example: Aquasar).
Modern CPUs get very hot during operation, depending on their workload. Depending on the model and manufacturer, power losses of up to 125 watts are achieved per square centimeter (current quad cores). For comparison: the 18 cm hotplate of a conventional electric cooker only achieves 7–10 W / cm².
Like all semiconductors , however, CPUs must not exceed certain operating temperatures, as this initially leads to malfunctions ("crashes") and in extreme cases to the destruction of the chip (this is prevented by overheating protection in newer processors). Usual limit temperatures for operation are between 60 and 90 ° C. Temperatures above around 125 to 135 ° C lead to irreversible damage. Processors must therefore be cooled, whereby a certain safety margin to the maximum values specified by the manufacturer is desirable.
The most common way to ensure the cooling of the CPU is to mount a heat sink with a fan. The ribbed heat sink made of aluminum or copper (partially combined) increases the area that contributes to the heat dissipation many times over, the fan is supposed to ensure that the heat loss is quickly removed. The cooling is often not dimensioned according to the theoretically maximum possible power loss, but for cost reasons according to the Thermal Design Power (TDP), which is significantly lower.
Thermal paste or a thermal pad is used between the processor and the heat sink . As a result of unevenness and roughness, air pockets remain between the chip and the heat sink, which impede the transport of heat ; the pastes or pads displace this air and improve the heat transfer considerably.
Axial fans with diameters between 40 and 140 mm are almost exclusively used as fans for the CPU cooler . Small specimens in particular reach speeds of up to 6500 rpm and can generate considerable background noise . The fans are now connected to the motherboard so that the speed can be monitored and, in many modern motherboards, can also be controlled electronically .
As an alternative to air cooling, there is also water cooling for powerful or relatively quiet computers, in which water is cooled inside or outside the computer in a radiator (sometimes without a fan) and then with the help of a pump through the housing and on objects to be cooled like CPU, sometimes also to RAM, chipset, graphics processor, etc. Overall, water cooling is more complex, expensive and usually more maintenance-intensive than air cooling. Apple was the first computer manufacturer to install standardized water cooling in its Power Mac G5 top models. Previously, water cooling was mostly only used by hobbyists with overclocked processors installed by themselves.
Liquid nitrogen cooling is also used in the industrial sector, but it is extremely complex. In order to be liquid, the nitrogen has to be cooled to −196 ° C, which requires large cooling units. Because of the very low temperature inside the computer, the motherboard and other objects must be reheated from the back in order for them to function properly. This technology is very difficult to implement, the operating and maintenance costs are usually higher than operating several individual processors in parallel. In general, it does not make sense to cool a CPU down to less than +10 ° C, otherwise the costs will be too high. All electronic components also have a minimum operating temperature and condensation can form on components that are too cooled , which must be avoided.
However, liquid nitrogen cooling makes sense as a short-term solution for setting new clock frequency and benchmark records. No cooling units are required for this either, the nitrogen is simply refilled from the bottle and evaporated. In this case, there is also no need to heat the back because the components usually remain functional during the short time required for a record attempt, even without such measures.
Individual manufacturers also use compression refrigeration machines . These work in a similar way to a refrigerator : a coolant is put under high pressure and the heat generated is dissipated; when it is equalized to normal pressure, it cools down further and thus also cools its surroundings, i.e. the processor or other devices. This solution is mainly used for overclocked workstations , but has the disadvantage of also generating the background noise of a refrigerator.
Another possibility for forced cooling of the CPU is the Peltier element . Here, too, there is a risk of condensation forming. In addition, due to its low efficiency , a Peltier element generates at least the same power loss as the processor itself, which must also be dissipated. The “warm” side has to be cooled by water cooling or a heat sink with a fan.
Heat can also be dissipated through the use of oil cooling, but in the PC area this has so far only been carried out in an experimental environment. In most cases, no special fans or cooling devices are installed on the CPU for oil cooling, but the entire motherboard with the fan installed is simply immersed in a tub full of oil. For this purpose, non-conductive, pure mineral oil is recommended .
Engineering samples / customer samples
The first CPUs produced by a manufacturer, similar to a prototype, are distributed to selected companies or testers as "engineering samples" or "confidential CPUs". Basically, these are fully functional processors that are usually in no way inferior to the later end product. Such CPUs are usually not commercially available. Such CPU versions can be recognized by the abbreviation "ES" or the imprint "Confidential". In addition, at least in the past from Intel, processors and entire chip sets were sold in "university kits". The chips contained there had the imprint "CS" and were usually damaged on the ceramic housing, in many cases the imprint was bad (slipped, smeared, duplicated). It should be noted that the letter combination ES or CS does not always have to mean engineering or customer sample; it is often also the batch code or a revision designation.
Intel Mobile Pentium 4 labeled Confidential .
- Thomas Flik: Microprocessor technology and computer structures. Springer-Verlag, Berlin / Heidelberg 2004, ISBN 3-540-22270-7 .
- Michael S. Malone: The microprocessor. An unusual biography. Springer-Verlag, Berlin / Heidelberg 1995, ISBN 3-540-60514-2 .
- Thomas Beierlein, Olaf Hagenbruch: Pocket book on microprocessor technology. 3. Edition. Fachbuchverlag Leipzig in Carl Hanser Verlag, Leipzig 2004, ISBN 3-446-22072-0 .
- Klaus Wüst: Microprocessor technology: basics, architectures and programming of microprocessors, microcontrollers and signal processors. Vieweg and Teubner, 2008, ISBN 978-3-8348-0461-7 .
- Dieter Wecker: Processor design. 2nd Edition. De Gruyter Oldenbourg-Verlag, 2015, ISBN 978-3-11-040296-4 .
- 25 microchips that shook the world . Institute of Electrical and Electronics Engineers - on the history of microprocessors
- Visualization of different addressing types ( Adobe Flash )
- Virtual Museum for CPU-images (English)
- Manuel Jiménez: Introduction to Embedded Systems. Springer Science & Business Media, 2013, ISBN 978-1-4614-3143-5 , p. 3 ( limited preview in Google book search).
- Ray M. Holt: World's First Microprocessor Chip Set. Ray M. Holt website, archived from the original on July 25, 2010 ; Retrieved July 25, 2010 .
- World's First Microprocessor. In: firstmicroprocessor.com. April 21, 1998, accessed August 9, 2015 .
- Jivan S. Parab, Vinod G. Shelake, Rajanish K. Kamat, Gourish M. Naik: Exploring C for Microcontrollers: A Hands on Approach . Springer, 2007, ISBN 978-1-4020-6067-0 , pp. 4 ( ee.sharif.edu [PDF]).
- From a Few Cores to Many: A Tera-scale Computing Research Overview. (PDF) IBM , 2006, accessed August 17, 2013 .