Fullscreen HTML5Med Res (??MB)HTML4

Image source: https://pixabay.com/photos/intel-8008-cpu-old-processor-3259173/ T o r y of t s i h he intel Pa rt 1 b y D r D avid Mad K3D V , n o d is SM Intel is (or some would say was) one of the world’s most influential and largest manufacturer of computer chips, including microprocessors. That includes the central processing units that power a large portion of modern computers and related devices. S tarting with the world’s first microprocessor in 1971, which sparked the personal computer revolution, Intel grew to a market capitalisation of US$509 billion in 2000 ($930 billion in today’s money). Today it sits at around US$188 billion and fluctuating, while facing AI challenges, serious competition and the legacy of management deficiencies, leading to failures to innovate, among other problems. Intel is currently building new foundries in the United States but still has management challenges after a rocky few years. The founding of Intel Fairchild Semiconductor was founded in 1957 by the “traitorous eight” engineers from Shockley Labs, who were dissatisfied with the way Shockley ran it. Two of those eight were Gordon Moore (famous for Moore’s Law) and Robert Noyce, the co-inventor of the integrated circuit (see Fig.1). Moore and Noyce left Fairchild to 16 Silicon Chip found Intel on the 18th of July 1968. Another Fairchild employee, Andy Grove, also left and joined Intel on the day of its incorporation, although he was not a founder. He helped get Intel’s manufacturing operations started and move Intel’s focus from memory to CPUs in the 1980s, establishing it as the dominant player in the market. In addition, investor Arthur Rock provided US$2.5 million in funding (equivalent to US$23.3 million today or AU$35.5 million). The new company was originally proposed to be named Moore Noyce, but they decided it was best to avoid the “more noise” pun, which is understandable for an electronics company. It was named NM Electronics initially, but after a few weeks, was renamed to Intel, which is derived from “integrated electronics”. Intel was already a trademark of the hotel chain Intelco, so they also had to buy the rights to that name. Intel’s first headquarters was in Mountain View, California (it is now in Santa Clara, California). Its first Australia's electronics magazine 106 employees are shown in Fig.2 (in 1969). Noyce and Moore left Fairchild because they saw the potential of integrated circuits (ICs) and wanted to create a company centred on their research and production. For more on Fairchild and the traitorous eight, see our articles on IC Fabrication in the June-August 2022 issues (siliconchip. au/Series/382). They had become dissatisfied at Fairchild because they felt it was not reinvesting enough in research and development. They felt Fairchild wasn’t growing enough, were dissatisfied with the administrative workload, and stated that it no longer had a hands-on creative culture like it used to have. They also wanted to standardise the mass-production of ICs. Specifically, what they wanted to standardise was a manufacturing process for chips that could be widely adopted, was cost effective, scalable and could be applied to many different chip designs. siliconchip.com.au Fig.1: Andy Grove, Robert Noyce and Gordon Moore in 1978. Source: www.flickr.com/photos/8267616249 (CC BY-SA 2.0) Noyce invented the first commercially viable monolithic IC (a circuit on a single piece of silicon or other material containing all the circuit’s transistors, resistors, capacitors etc) and licensed Fairchild’s “planar process” for manufacturing it. Thus, the new company was to be based on investing extensively in research into the manufacturing of integrated circuits, with a focus on standardisation of the production processes for the monolithic ICs. Moore’s Law provided an ongoing objective for Intel to strive toward. Moore’s Law was an observation he made in 1965 that the number of components on a chip doubles roughly every two years, a compound growth rate of 41%. Moore’s Law held until roughly 2016, at which time the physical limits of component density were reached. The rapid increase in computing power continues through advanced chip packaging methods, architectures and higher clock speeds. Intel’s striving to fulfil Moore’s Law siliconchip.com.au Fig.2: a photo of Intel’s first 106 employees in 1969. Source: https://intelalumni.org/memorylane allowed for an ongoing reduction in the cost of ICs and computers to consumers. That’s because fitting more components onto one silicon chip means a more powerful device for the same cost or less. Conversely, the cost to producers, including Intel, to continue to manufacture higher and higher component densities increases as it becomes more difficult to make cheaper and faster chips. The hope is that improvements in manufacturing technology and economies of scale reduce the cost enough that chips become both more powerful and also cheaper. Intel processor history overview Intel is mostly identified with lines of microprocessors, although it has created many other products, which we will also discuss. Since Intel has produced such a wide range of processors, its history is complicated and can be hard to follow. An abbreviated timeline of Intel processor release dates is shown in Table Australia's electronics magazine 1 overleaf. Many of these will be discussed in more detail later. Understanding Intel’s history Intel has a complex history, so we have broken it up into its dominant features in every decade. The main features of each decade can be summarised as follows: 1970s invented the microprocessor almost by accident with the 4004; the 8080 derivative launched the microprocessor revolution. 1980s Intel dominated the establishment of the PC era. The IBM PC was released, using the 8088, 80286, 80386 or 80486. Along with clones, it became the dominant PC. 1990s Intel continued to dominate the PC market. Intel and Pentium became household names, helped by the “Intel Inside” advertising campaign. 2000s the NetBurst architecture ultimately failed, losing market share to AMD, which reached 25% in 2006. They clawed back some ground with the Core microarchitecture February 2026  17 diversification, but faced various challenges. 2010s stagnation, delays in the 10nm process node, mobile market failure, AMD catching up. 2020s Taiwan Semiconductor Manufacturing Company (TSMC) technologically overtook Intel. Despite this, Intel still has foundry ambitions and developed hybrid cores. Unlike TSMC, Intel is an integrated device manufacturer (IDM) that designs, manufactures and sells its own chips; Intel wants to become the TSMC of the West. The IDM 2.0 strategy of CEO Pat Gelsinger saw five nodes in four years from 2021 to 2025: Intel 7, Intel 4, Intel 3, Intel 20A and Intel 18A. Now that we’ve given a broad overview, let’s look at Intel’s history in more detail. Table 1: Intel processor families Processor family Release date 4004 1971 8086/8088 1978 80286 1982 80386 1985 80486 1989 Pentium 1993 Pentium Pro 1995 Pentium II 1997 Pentium III 1999 Pentium 4 2000 Core & Core 2 2006 Core i3/i5/i7 (1st-8th gen) 2008-2017 Core i3/i5/i7/i9 (9th-14th gen) 2018-2023 1969-1970s: starting as a memory company Core Series 1 2023 Intel began the decade as the world’s leading memory chip maker and ended it by accidentally igniting the personal computer revolution with the 4004 (1971) and then the 8080 (1974). The 4004 microprocessor was originally just a side project for calculators, but became the company’s future when dynamic random access memory (DRAM) profit margins started to collapse. Core Series 2 2024-2025 Core Series 3 Early 2026 Intel’s first products Intel’s most important early products, which established the microcomputer revolution, were based around five chips or chipsets. These were the 3101 (memory), 1101 (memory), 1103 (memory), 1702 (EPROM or erasable programmable read-only memory) and the 4004 (microprocessor) and its associated chipset. We will now describe each of these chips. 1969: Intel 3101 Intel’s first product was the 3101 Schottky TTL bipolar 64-bit static random access memory (SRAM) chip, released in April 1969. By today’s standards, it had an incredibly small storage capacity, equivalent to just eight characters (64 bits). Nevertheless, it was a remarkable achievement as the company was only established in July 1968. Due to the use of Schottky technology, it was nearly twice as fast as earlier implementations of such chips and was designed for use with computer CPUs. Even though Intel initially wanted to focus on research and development, they were incentivised to produce this chip by Honeywell’s announcement that they would purchase SRAMs from anyone who made them. This triggered a competition among memory manufacturers. Honeywell ended up not using the chips because they wanted more than 64 bits, but Intel’s achievement made it known to the world that Intel was now a serious company, no longer the underdog, and other companies became interested in the 3101. The 3101 was unsuitable for main memory, the dominant form of which at the time was magnetic core memory, which had capacities in mainframes up to around 4MiB (in the IBM 360 model 195). Still, it was suitable where high-speed memory devices were needed, such as for processor registers in minicomputers as offered by Burroughs, Xerox and Interdata. 1969: Intel 1101 Following soon after the 3101 was an even more important product, the 1101 256-bit SRAM chip (Fig.3), which was the first with two key technologies: metal oxide semiconductor (MOS) and silicon gates rather than metal. The MOS technology allowed for higher memory capacity (more memory per area of silicon) and higher chip densities. It had access times of 1.5 microseconds (1.5μs) and ran at 5V, consuming 500mW. 1970: Intel 1103 The 1103 (Fig.4) was the first commercial DRAM (dynamic random access memory) memory chip with a The difference between SRAM and DRAM SRAM is faster than DRAM while using less power, as it doesn’t need constant refreshing to maintain data, but it is more expensive and has a lower capacity per chip than DRAM. On the flip side, DRAM is cheaper and has a higher capacity per chip, but it uses more power and is slower than SRAM as it needs to be constantly refreshed. Both types of memory are volatile, meaning they lose their data when power is removed. Fig.3 (top): Intel’s first really successful product, the 1101 256-bit SRAM chip. Source: www. cpu-zone.com/1101.htm Fig.4 (bottom): Intel’s first DRAM chip, the 1103 introduced in 1970. Source: https://w.wiki/ GYXb (CC BY-SA 4.0) 18 Silicon Chip Australia's electronics magazine Fig.5: the three-transistor memory cell was invented in 1969 by William Regitz and colleagues at Honeywell. Original source: https://w.wiki/GYJp (GNU FDL v1.2) siliconchip.com.au capacity of 1024 bits or 128 extended ASCII characters. It had a sufficiently high capacity and low enough cost that it began to replace magnetic core memory. By 1972, it was outselling all other types of memory combined due to costing less and being smaller than core memory. The chip was discontinued in 1979. It was used in computers such as the HP 9800 series, Honeywell minicomputers and the PDP-11. The actual three-transistor dynamic memory cell configuration shown in Fig.5 was invented by Honeywell, who asked the fledgling Intel to manufacture it. It was later also manufactured by National Semiconductor, Signetics and Synertek. 1971: Intel 1702 The first EPROM chip was developed by Dov Frohman at Intel – see Figs.6 & 7. It had 2048 bits of memory that could be erased with UV light and rewritten electrically. It was revolutionary because, before then, “firmware”, the most basic instructions for a computer or similar device to boot, had to be in the form of hardwired logic that was difficult or impossible to change. Intel offered another cheaper version of this chip, which was ‘write once’ and could not be erased. The only differences were that it did not have an expensive transparent quartz window for UV erasure, and it came in a plastic rather than ceramic package. Today, flash memory has replaced EPROM memory for things like firmware, but the 1702 was an important development as it made prototyping new products much easier, along with allowing product updates. Fig.6: a demonstration of the 1702 chip in 1971, using its stored information to display the Intel logo on an oscilloscope. Source: https:// timeline.intel.com/1971/the-world’sfirst-eprom:-the-1702 Fig.7: the Intel 1702 had a transparent window through which the contents could be erased by UV light and then electronically rewritten. Source: https://timeline.intel.com/1971/theworld’s-first-eprom:-the-1702 1970s: the microprocessor revolution Intel’s and the world’s first microprocessor would not have happened at the time had it not been for a request from the Japanese Busicom calculator company. The Busicom calculator In 1969, Busicom asked Intel to design a set of chips for their proposed electronic calculator. At the time, calculators contained large numbers of discrete components and complex wiring, so they wanted to reduce the cost by using a dedicated chipset. The siliconchip.com.au Fig.8: a Busicom 141-PF / NCR 18-36 circuit board with chips Intel developed for it. Note the blank space for the optional 4001 ROM for the square root function. Source: Nigel Tout, http://vintagecalculators.com Busicom engineers designed a calculator that required 12 ICs and asked Intel to make these custom chips. Ted Hoff at Intel, aided by Federico Faggin and Stanley Mazor, came up with a much more elegant design needing only four chip types Australia's electronics magazine containing ROM (read-only memory), RAM (random-­access memory), a shift register and what was to become the 4004 microprocessor. These chips were developed, produced and sent to Busicom in January 1971, and they had exclusive rights to them. February 2026  19 The 4004 microprocessor was a single silicon chip that contained all the basic functional elements of a computer’s central processing unit (CPU). Until the 4004, CPUs had to be fabricated using multiple individual components at much greater cost and complexity. The resulting calculator was the Busicom 141-PF, also marketed as the NCR 18-36 (see Fig.8). An optional ROM chip was available to provide a square root function. In common with other calculators of the era, it printed the results rather than displaying them on a screen. This was an important moment in the history of calculators because, at the time, calculators had to have their functionality designed into hardware, which meant every calculator required extensive customised hardware. The new Intel microprocessor and ROM allowed new designs to be made simply by changing the programming of the microprocessor via ROM. The calculator used four 4001 ROM chips, two 4002 RAM chips, three 4003 shift registers and one 4004 microprocessor. More about this chipset later. At the same time as the Intel developments, Busicom commissioned Mostek to produce a ‘calculator on a chip’, which resulted in an even lower chip count than the Intel solution. The chip developed and released in November 1970 was the Mostek MK6010, but that’s another story. In mid-1971, Busicom asked Intel to lower the chip prices, which resulted in Intel renegotiating the contract such that Busicom gave up their exclusive rights, enabling Intel to sell the chips. Then, in November 1971, Intel announced the release of the MCS-4 chipset family based on the chips developed for Busicom. 1971: the beginning of the microprocessor revolution On the 15th of November 1971, Intel commercially released the 4004 microprocessor that they had developed for Busicom and licensed back to themselves. The Intel 4004 was a revolutionary product for the computer industry. It was designed to be affordable, easy-touse and accessible to a wide variety of computer designers. Early microprocessors such as the 4004 were not initially intended for general-purpose computing, but to run embedded systems such as calculators, cash registers, computer games, computer terminals, industrial robots, scientific instruments etc. In addition to the Busicom calculator mentioned above, it was used in Busicom automated teller and cash machines, the Intellec 4 microcomputer from Intel (Fig.9) to support software development for the 4004, a prototype pinball machine by Bally, and the Pioneer 10 spacecraft. The software to run such systems could be developed on the Intellec 4 and then permanently programmed into ROMs such as the 4001 during manufacture, or burned into EPROMs such as the 1702 (which could be erased and updated). The 4004 cost US$60 at the time, which in today’s money would be US$501 or AU$774. The MCS-4 (see Fig.9: the Intellec 4 microcomputer for software development for the 4004, available to developers only. It was programmed via front panel switches or an optional terminal interface. Source: https://w.wiki/GYJr (CC BY-SA 3.0) 20 Silicon Chip Australia's electronics magazine Fig.10) included the 4001 ROM, 4002 RAM and 4003 I/O chips that together formed the basic elements of a complete computer. The ~$750 price is similar to that of a high-end (consumer) CPU today. The 4004 contained 2300 transistors and was fabricated using a 10-micron (10μm) process. It could execute 60,000 instructions per second with a 740kHz clock speed and a 4-bit architecture. It could address 640 bytes of RAM and up to 4kiB of ROM – see Fig.11. The specifications of the MCS-4 chipset chips were: 4001 a 256 × 8-bit (256 byte) ROM. 4002 a 4 × 20 × 4-bit (40 byte) DRAM. 4003 an I/O chip with a 10-bit static shift register, serial and parallel outputs. A static shift register comprises flip-flops that store and shift binary data. 4004 the microprocessor. Using this chipset, a fully expanded 4004 system using sixteen 4001s could have 4kiB of ROM and sixteen 4002s for a total of 640 bytes of RAM, plus an unlimited number of 4003s for I/O. The most powerful 4004 system? The most powerful Intel 4004 system, called Linux/4004, was built by Dmitry Grinberg in 2024. It was created to use “ancient” 4004 hardware merged with a modern Linux operating system. It is a testament to the powerful and flexible nature of the 4004 chip, which was originally intended to power a calculator, but is not exactly practical. The system took 4.76 days to boot a stripped-down Linux kernel to the Fig.10: the Intel MCS-4 chipset. Source: https://en.wikichip.org/wiki/ File:MCS-4.jpg siliconchip.com.au Fig.11: the chip layout (a drawing, not a photograph) of the 4004 processor. Source: https://w.wiki/ GYJq (CC0 1.0) 4004 image source: https://w.wiki/GYZY 8008 image source: https://w.wiki/GYZZ i960 image source: https://w.wiki/GYK8 Fig.12: the die of the Intel 8008, their first 8-bit CPU. Source: https://x.com/ duke_cpu/status/ 1980293005644107812 Fig.13: an Intel i960 die (80960JA). Note the large cache memory banks (rectangular grids); the actual core is pretty small since it’s a RISC processor. Source: https://w.wiki/GYK9 (CC BY 3.0) siliconchip.com.au Australia's electronics magazine February 2026  21 Fig.14: an 8080 chip made by Intel. Source: https://w.wiki/GYJy (CC BY 4.0) Fig.15: the Altair 8800 computer was sold as a kit, and also has an optional 8-inch floppy drive. It popularised the use of the Intel 8080 processor. Source: https://americanhistory.si.edu/collections/ object/nmah_334396 command prompt. It could perform rudimentary mathematical fractal calculations of the Mandelbrot set. A full description of the project can be found at siliconchip.au/link/ac9t and there is a video on it at https://youtu. be/NQZZ21WZZr0 After the 4004 The success of the 4004 led to the development of the 8008 and the 8080 CPUs, which established Intel as the world leader and led to great expansion of the company in the 1970s, 1980s and 1990s. 8008 the 4004 led to the development of the 8008 in April 1972. It was the first 8-bit microprocessor and could address 16kiB of memory. It was manufactured but not designed by Intel. CTC (Computer Terminal Corporation) designed it for use in their Datapoint 2200 programmable terminal, but Intel licensed the design for use in other products. The 8008 was discontinued in 1983. Its clock speed was 500-800kHz and it used 10-micron technology, with 3500 transistors. The 8008 is most famous for being the microprocessor used in the first enthusiast personal computers: the SCELBI (US, 1974), the Micral N (France, 1973) and the MCM/70 (Canada, 1973). It was also used in the HP 2640 computer terminals. 8080 the 8080 followed in 1974 (Fig.14). It was originally conceived for embedded systems, but it was broadly adopted and remained in production until 1990. Made with a 6 micron (6μm) process node, it had a clock rate of 2-3.125MHz and was an 8-bit processor but had the ability to execute 16-bit instructions. It could address 64kiB of memory. 22 Silicon Chip A variety of support chips were available for it. It had about 6000 transistors and could execute several hundred thousand instructions per second. It was used in the first commercially successful personal computers, like the Altair 8800 (see Fig.15), and other S-100 bus systems running the CP/M operating system. 8085 the 8085, introduced in March 1976 and discontinued in 2000, was the successor to the 8080 and Intel’s last 8-bit processor. It was compatible with the 8080 but had the advantage of only needing to be supplied with one voltage, not three like the 8080, making system development simpler. It ran at a clock speed of 3MHz, 5MHz or 6MHz, used a 3 micron process node and had 6500 transistors. It was not widely adopted in microcomputers because the Zilog Z80 chip (1976-2024) was introduced, which took over much of the 8-bit market (eg, running the Osborne 1, TRS-80 and ZX Spectrum). However, the 8085 was used as a microcontroller and in video terminals like the VT-102. 8086 in 1978, Intel introduced the 8086, its first 16-bit processor with 29,000 transistors, built on a 3.5 micron process (switching to 2 microns in 1981) – see Fig.16. It extended the 8080 architecture, introduced segmented memory addressing, ran at up to 10MHz and could support 1MiB of RAM. It had a simple two-stage pipelining unit to improve performance. It laid the foundation of the x86 instruction set family of processors. This processor, along with dominance of the memory chip market, paved the way for the commercial personal computer boom. The x86 instruction set The x86 instruction set that’s still widely used today was introduced with the 8086. It became standardised with the release of the 8088 processor thanks to its use by IBM in their open PC architecture in 1981. x86 has had many updates over the years, but today’s processors can still run code that was written back in the late 1970s. This does not mean that such code will run on a modern operating system like Windows 11, but that is a restriction of Windows, not the processor itself. It is possible to boot Microsoft DOS from 1981 on a current x86 CPU. There would be problems such as a lack of USB and other driver support, and a lack of compatibility with a modern UEFI (unified extensible firmware interface) BIOS. There is a video of a system with a 2016 Intel Celeron N3450 CPU booting a 45-year-old version of DOS at https://youtu.be/BXNHHUmVZh8 (the Celeron name was generally applied to a cut-down or simplified Pentium processor). Microsoft also played a role in the standardisation of x86 by supporting a wide range of hardware that used x86. With time, new instructions have been added Fig.16: an 8086 chip in a ceramic dual-inline package (DIP). Source: https://w.wiki/GYK4 (CC BY-SA 4.0) Australia's electronics magazine siliconchip.com.au to x86, but the old ones have been kept to ensure compatibility. Intel and AMD, who both make x86-compatible processors, have formed an alliance to standardise future instructions to ensure their consistent implementation across future products from both companies. Competing instruction sets include ARM, MIPS and RISC-V. Backward compatibility is important because there are enormous amounts of commercial, financial, industrial, military, medical and domestic software written for old processors that may still be in use. Some of this software, which can be decades old, runs on DOS, including accounting software, payroll systems, programmable logic controllers, CNC machines and retro games. This is one reason that attempts to replace the x86 instruction set have not generally been successful, although ARM has made some inroads. Emulation (where software running on one processor can interpret instructions from a different set) can help to ease the transition. From 2020 to 2023, Apple moved away from the x86 architecture as they transitioned from Intel microprocessors (which they used since 2006) to their own designs based on the ARM architecture. Apple’s reasons were they wanted a common technology across all their platforms, better performance per watt and they wanted to integrate all components on a single chip (see also the section later on the stagnation of Intel’s innovation). Over the years, Intel has developed extensions to the x86 instruction set, including: ● MultiMedia eXtensions (MMX) ● the Streaming SIMD (single instruction, multiple data) Extensions, which superseded MMX: SSE, SSE2, SSE3 and SSE4 ● Advanced Vector eXtensions (AVX, AVX2 and AVX-512) ● Advanced Encryption Standard – New Instructions (AES-NI) ● Software Guard eXtensions (SGX) ● Trusted eXecution Technology (TXT) ● Transactional Synchronisation eXtensions (TSX) ● haRDware RANDom number generator (RDRAND) ● Carry-Less MULtiplication for cryptography (CLMUL) siliconchip.com.au Table 2 – Intel’s process node names (only consumer CPUs listed) Year Process Name Chips made # transistors 1972 10μm 10μm 4004 2.3k 1974 8μm 10μm 4040 3k 1976 6μm 6μm 8080 6k 1977 3μm 3μm 8085, 8086, 8088 29k 1979 2μm 2μm 80186 134k 1982 1.5μm 1.5μm 80286, 80386 275k 1987 1μm 1μm 80386, 80486 (up to 33MHz) 1.2M 1989 800nm 800nm 80486 (up to 100MHz) 1.3M 1991 600nm 600nm 80486 (100MHz), Pentium (60200MHz) 3.1M 1995 350nm 350nm Pentium (120-200MHz), Pentium MMX (166-233MHz), Pentium Pro (150-200MHz) 5.5M 1997 250nm 250nm Pentium Pro, Pentium II (233450MHz), Pentium III (450600MHz) 9.5M 1999 180nm 180nm Pentium III (500-1133MHz), Pentium 4 (NetBurst, 1.3-1.8GHz) 42M 2001 130nm 130nm Pentium III (1.0-1.4GHz), Pentium 4 (NetBurst, 1.6-3.4GHz) 125M 2003 90nm 90nm Pentium 4 (NetBurst, 2.4-3.8GHz), Pentium M 169M 2005 65nm 65nm Final Pentium 4, Core, early Core 2 Solo / Duo 291M 2007 45nm 45nm Late Core 2 Duo / Quad, Core i3/ i5/i7 (1st gen) 731M 2009 32nm 32nm Core i3/i5/i7 (1st gen refresh & 2nd gen) 1.17B 2011 22nm 22nm Core i3/i5/i7 (3rd & 4th gen) 1.4B 2014 14nm 14nm Core i3/i5/i7/i9 (5th to 9th gen) 3B 2019 10nm 10nm Core i3/i5/i7/i9 (10th & 11th gen) 4.1B 2021 10nm+ Intel 7 Core i3/i5/i7/i9 (12th & 13th gen) 21B 2023 5nm Intel 4 & 3 Core i3/i5/i7/i9 (14th gen), Core Ultra 1 30B 2024 3nm Intel 20A Core Ultra 2 ~45B 2025 2nm Intel 18A & 14A Core Ultra 3 ~80B ● x86-64, a 64-bit version of x86 that allows, among other things, access to more than 4GB of RAM (developed by AMD but also implemented by Intel) ● Advanced Performance eXtensions (APX) Process nodes Throughout Intel’s history, it was shrinking the feature size of its chips, achieving higher numbers of transistors and higher component densities. We will divert from the history for a moment to describe process nodes, an essential part of understanding subsequent processor development. Australia's electronics magazine A process node (or technology node, which means the same thing) is a term used in semiconductor manufacturing representing a specific generation of chip technology. It was traditionally named based on the size of a transistor gate, which continued to shrink while Moore’s Law still applied. As it is difficult to shrink transistors much more than they are now, the names no longer correspond to any particular physical size, and are more of a marketing term representing performance and density increases, which continue due to 3D packaging and other techniques. February 2026  23 Intel 4004 Architecture D0-D3 bidirectional Data Bus Data Bus Buffer 4 Bit internal Data Bus Temp. Register Register Multiplexer Instruction Register Stack Multiplexer Flag Flip Flops ALU Stack Pointer Instruction Decoder and Machine Cycle Encoding Index Register Select Accumulator Program Counter Level No. 1 Level No. 2 Level No. 3 Address Stack Decimal Adjust 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Scratch Pad Timing and Control ROM Control RAM Control Test Sync Clocks CM ROM CM RAM 0-3 Test Sync Ph1 Ph2 Reset Fig.17: the microarchitecture of Intel’s (and the world’s) first microprocessor, the 4004 from 1971. Source: https://w.wiki/GYJu (GNU FDL v1.2) 32 KB Instruction Cache (8 way) 128 Entry ITLB Shared Bus Interface Unit 128 Bit 32 Byte Pre-Decode, Fetch Buffer Instruction Fetch Unit 6 Instructions 18 Entry Instruction Queue Complex Decoder Microcode Simple Decoder Simple Decoder 4 µops 1 µop Simple Decoder 1 µop 1 µop Shared L2 Cache (16 way) 7+ Entry µop Buffer 4 µops Register Alias Table and Allocator 4 µops 4 µops 96 Entry Reorder Buffer (ROB) Retirement Register File (Program Visible State) 256 Entry L2 DTLB 4 µops 32 Entry Reservation Station Port 0 ALU SSE Shuffle ALU ALU 128 Bit FMUL FDIV Intel Core 2 Architecture SSE Shuffle MUL 128 Bit FADD Internal Results Bus Port 3 Port 5 Port 1 ALU Branch SSE ALU Store Address Port 4 Store Data Port 2 Load Address Memory Ordering Buffer (MOB) 128 Bit Store 128 Bit 32 KB Dual Ported Data Cache (8 way) Load 256 Bit 16 Entry DTLB Fig.18: the microarchitecture of the much more advanced Intel Core 2 processor from 2006. Source: https://w.wiki/GYJv (GNU FDL v1.2) 24 Silicon Chip Australia's electronics magazine The number of atoms across the smallest dimension of a transistor of the Intel 18A process node (representing 18Å or 1.8nm) is estimated to be 180, but because of the 3D nature of the transistor, the overall number is estimated to be thousands. This is currently the minimum number required for reliable function. That might not be improved on for a long time, if ever, for practical devices as adverse quantum mechanical effects like electron tunnelling are already a concern with the 18A process node. But technology always develops in unexpected ways... By way of comparison, the smallest process node described by Samsung is 2nm or 20Å. The distance between centres of silicon atoms in a crystal lattice is 0.235nm or 2.35Å. Commonly used terms for Intel fabrication processes are listed in Table 2. The 18A process node (1.8nm) is what Intel is focusing on for the future. It will be produced at its Arizona and Oregon foundries, which are its most advanced in the world and will lead the way to the “one trillion transistor laptop”. This process node incorporates all the above technologies and is the culmination of the so-called 5N4Y (five nodes in four years), which was former CEO Patrick Gelsinger’s turnaround strategy, announced in 2021. Gelsinger was asked to leave in December 2024 when the board felt improvements were not being made fast enough (his replacement has had some controversies). The 5N4Y plan nodes were: Intel 7 (~10nm): the first use of their Enhanced SuperFin transistors. Intel 4 (~5nm): produced with extreme ultraviolet (EUV) lithography and moving to chiplets/tiles and associated interconnect technologies, like Foveros and EMIB (more on these later). Intel 3 (~5nm): with improved performance per watt. Intel 20A: A marks the move to Angstrom-­ b ased measurements. It didn’t go into full production, but led the way to the implementation of Ribbon FETs and PowerVias in 18A (more on these later). Intel 18A: the current process node with the first processor being the Core Ultra series 3 (Panther Lake) and the second to be the Xeon 6+ (Clearwater Forest). siliconchip.com.au Microarchitectures Microarchitecture (or μarch) is the particular way a processor’s internal hardware (pipelines, execution units, caches etc) is designed and organised to implement a given instruction-set architecture (ISA) such as x86. It is typically illustrated with pipeline or block diagrams, like Figs.17 & 18. Intel re-uses microarchitectures across multiple processor generations and models. Most (but not all) major new Intel processor families introduce a new or significantly revised microarchitecture. A new microarchitecture appeared every 2-4 years, while new processor series (new brand names or model numbers) were released every 12-18 months; this was called their tick-tock model. Examples of Intel microarchitectures are shown in Table 3. Let’s now look at more recent eras of Intel products. The 1970s PC boom Intel’s processors of the 1970s had a great cultural impact and were a leap forward for microcomputing via the hobbyist PC boom of that era. They were responsible for democratising computing and sparking a global DIY computer revolution, which ultimately led to the widespread commercial development of microcomputers. As mentioned, the 8080 was released in 1974. It was the first truly affordable 8-bit CPU that a hobbyist could purchase. It cost US$360 in single units, but kit manufacturers like MITS, the creators of the Altair 8800, could get them for US$75 (equivalent to AU$757 today) in volume and sell them via mail order. The chip was small, relatively inexpensive and well-documented, so it was something hobbyists could make something with. Thus, computing moved out of the corporate lab and into garages and bedrooms. The Altair 8800 featured on the cover of Popular Electronics magazine in 1975 and, after that, 4000 were sold in weeks at US$439 (AU$4000 today) pre-assembled or US$297 (AU$2750 today) as a kit. Hobbyists saw the chip and the Altair computer that used it as a ‘blank canvas’. After seeing the magazine, Bill Gates and Paul Allen wrote Altair BASIC in 1975 as Microsoft’s (then called MicroSoft) first product. They used a PDP10 mainframe running an 8080 emulator. Gates released the source code siliconchip.com.au Table 3 – Intel microarchitectures from 1993 to the present Microarchitecture Years Processor families or brands P5 1993-1997 Pentium (60–200 MHz), Pentium MMX P6 1995-2003 Pentium Pro, Pentium II, Pentium III, Celeron (early), Pentium II Xeon, Pentium III Xeon NetBurst 2000-2007 Pentium 4, Pentium D, early Xeon Core 2006-2008 Core 2 Duo / Quad (Yonah → Penryn) Nehalem 2008-2010 Core i3/i5/i7 (1st gen) Sandy Bridge 2011-2012 Core i3/i5/i7 (2nd & 3rd gen) Ivy Bridge 2012-2013 3xxx series (22nm shrink + tweaks) Haswell → Broadwell → Skylake → … → Coffee Lake → Comet Lake → Rocket Lake 2013-2021 4th gen → 11th gen Core (various), Skylake derivatives used for six consecutive generations (2015-2021) Alder Lake (Golden Cove + 2021-2023 Gracemont cores), Raptor Lake 12th, 13th & 14th gen Core Meteor Lake 2023 Series 1, chiplet-based design Arrow Lake / Lunar Lake 2024-2025+ Series 2 (15th Gen) in April 2025 to mark Microsoft’s 50th anniversary. Steve Wozniak was also inspired by the Altair, which motivated him to design his own computer, the Apple I kit, released in July 1976. It used fewer parts than the Altair. He demonstrated it at the Homebrew Computer Club and shared the design and software for free, but the basic kit was sold for US$666.66 or AU$5800 today. It did not use an Intel processor, but a MOS 6502 instead. The Homebrew Computer Club held Silicon Valley garage meetings where hobbyists shared 8080 designs and code. Intel provided free datasheets, reference designs and even engineers who attended. Their slogan was “Build it. Share it. Improve it.” Other hobbyist computers of the 1976-1979 era were the IMSAI 8080, with the Intel 8080, and computers inspired by the 8080, like the TRS-80 (1977) that used the Zilog Z80 (which was 8080 compatible), and the Commodore PET (1977), which used the MOS 6502 like the Apple I. Intel provided open documentation for its products and encouraged chips such as the Z80 which, being compatible with the 8080, helped establish the 8080 ecosystem. This led to the dominant x86 architecture, which is still in widespread use today. Hobbyist computer magazines supported this new technology; magazines like BYTE, Creative Computing, Kilobaud Microcomputing and Dr Dobb’s Journal. During this period, there were price drops of the 8080, 8085 and 8088 chips, which led to mass adoption of microprocessors. By 1980, hundreds of thousands of hobbyists worldwide were programming in assembly language, swapping floppies and “building the future”. In 1978, Intel released the first electrically erasable programmable readonly memory (EEPROM), the 2816, which had a capacity of 16kib (2kiB). It is non-volatile, meaning it retains its memory when the power is switched off, but it can be erased and rewritten when desired without needing a UV light source, as the earlier 1702 EPROM did. It is considered a major achievement in the history of computing, allowing easy in-system reprogramming for both hobbyists and commercial users. The IBM PC is introduced In 1979, the 16-bit 8088 CPU with 29,000 transistors was introduced as a lower-cost version of the 8086 (see Fig.19). It was the heart of the original Fig.19: an original Intel 8088 processor. Source: https://w.wiki/GYJw (CC BY-SA 4.0) Australia's electronics magazine February 2026  25 IBM Personal Computer, which was released on the 12th of August 1981 (see Fig.20). Even though it was a 16-bit processor, external communications were via an 8-bit data bus for cost efficiency, but it could address 1MiB of memory with its 20-bit memory address bus. It was designed in Israel (as many of Intel’s processors have been). IBM’s decision to use the 8088 led to the standardisation of the x86 instruction set, because IBM’s open architecture approach encouraged cloning of the computer and the development of compatible expansion cards, which led to the rapid expansion of the Intel and x86 ecosystem. Also, IBM insisted on a second source for their PC chips, leading to Intel licensing their designs to AMD. AMD continues to make Intel-­ compatible CPUs to this day. It had simple pipelining in the form of a prefetch queue that read instructions from memory before they were needed. This enabled a performance increase. An 8087 mathematical coprocessor was available to complement the 8086 or 8088, which dramatically improved the speed of floating-­ point arithmetic operations. 1980s: dominating the PC era A low point of the 1980s for Intel was being forced out of the DRAM market by Japanese competition. Intel’s DRAM market share had fallen from over 80% in the 1970s to 2-3% by 1985, and they decided to withdraw from the market and fully focus on microprocessors. Intel bet everything on the x86 family. The 80386 (1985), in particular, turned the IBM PC standard into a near-monopoly and made Intel the indispensable heart of personal computers. The IBM PC and its clones dominated the PC market and cemented the legacy of the x86 instruction set that is used in almost all Intel and many competing processors (eg, from AMD) to this day. By the end of the decade, x86 processors generated almost all the company’s profit, and Intel processors dominated the PC market. Other processors they developed in this era were: iAPX 432 The iAPX 432 (1981-1985) was Intel’s ambitious but ultimately unsuccessful first attempt at a true 32-bit microprocessor. It comprised two chips (the 43201 general data processor and 43202 interface processor), was not based on the x86 architecture, and represented a radical departure from Intel’s prior designs. The 432 was designed from the ground up to support high-level languages like Ada directly in hardware, with features like object-oriented memory management, ‘garbage collection’ (a means to manage and recover unused memory) and capability-­based addressing (a memory and resources access model in which access is granted via tokens rather than raw addresses). These ideas were decades ahead of their time. This allowed modern operating systems to be implemented with significantly less code. However, technological limitations resulted in a performance roughly one-quarter that of the 80286, despite its advanced architecture. Compounding the problem, the 432 was not backward compatible with any existing Intel processor, alienating developers accustomed to the 8086/8088 ecosystem. These factors, combined with its high cost and complexity, led to its commercial failure. Fig.20: the original IBM PC from 1981, built around the Intel 8088. Source: https://w.wiki/ GYJx (CC BY-SA 3.0) 26 Australia's electronics magazine 80286 The 16-bit 80286 microprocessor was introduced in 1982 (Fig.21). It added ‘protected mode’ operation, enabling it to address up to 16MiB of memory instead of the 1MiB of the 8088, with improved multitasking capabilities compared to the ‘real mode’ limitations of earlier x86 chips. 16-bit data could be fetched in one bus cycle, while the 8088 required two bus cycles. Clock speeds up to 20MHz were supported, and the ‘286 facilitated more advanced operating systems such as IBM’s OS/2, Windows 3.0, Concurrent DOS, Minix and QNX that supported multitasking and more memory access compared to standard DOS. A disadvantage of ‘286 protected mode was that there was no way to return to real mode without a CPU reset, so standard DOS programs could not be run once the CPU was switched to protected mode. The ‘286 had simple pipelining, allowing the instruction unit, address unit, bus unit and execution unit to work concurrently to improve performance. An 80287 mathematics coprocessor was available. The ‘286 had between 120,000 and 134,000 transistors depending upon the variant, and was built using a 1500nm (1.5μm) process. The direct competitor to the ‘286 was Motorola’s 68000 (“68k”), which was used in the first Apple Macintosh, Commodore Amiga and Atari ST. It was a 32-bit processor with a 16-bit bus, but the ‘286 gave superior real-world benchmarks, and the IBM PC had an open architecture, giving it more software compatibility and therefore more popularity than the 68000. 80386 The 80386 was released in 1985, and came in two versions: the lower-­ priced SX, with a 32-bit internal architecture but a 16-bit external data bus and 24-bit memory address bus; and the DX, which was the ‘full’ version with a 32-bit external bus (Fig.22). It could support up to 4GiB of physical memory and up to 64TiB of virtual memory using advanced segmentation and paging. It was designed specifically with multitasking in mind. It had a simple six-stage instruction pipeline to allow the execution of different phases of certain instructions somewhat in parallel over multiple clock cycles, to siliconchip.com.au Fig.21: an 80286 chip. Source: https://w.wiki/ GYK6 (CC BY-SA 3.0) keep the processor busy at all times. Mathematical co-processors (80387) were available for both versions of the ‘386. It had 275,000 transistors and was built with a 1000nm (1μm) process. A special version produced for IBM, the 386SLC, had a large amount of on-chip cache, with 855,000 transistors. i960 Intel’s i960 (also known as the 80960), sold over 1988-2007, was a major shift away from the x86 architecture toward RISC (reduced instruction set computer) principles, which streamlines the instruction set, theoretically enabling faster execution – see Fig.13. It was mainly used as an embedded processor in military, industrial, and networking systems and achieved great success in niche markets such as laser printers, routers and even the F-22 Raptor stealth fighter. Intel discontinued the i960 in 2007 as part of a legal settlement with Digital Equipment Corporation (DEC) over patent disputes. In exchange, Intel gained rights to DEC’s Strong­ ARM design. 80486 The 80486 (Fig.23) was introduced in 1989. It had a built-in floating-point unit and so did not need an external coprocessor. It also had an inbuilt 8kiB cache, later increased to 16kiB in the DX4 variant, which gave it much better performance compared to the ‘386. It also had a five-stage pipelining architecture, similar to the ‘386 but with a more advanced architecture. Even though the 8088, 8086, ‘286 and ‘386 had instruction pipelining, the ‘486 was the first in which pipelining was tightly integrated. The 486SL variant was optimised for lower power consumption in laptops. It had 1.2-1.6 million transistors dependent upon variant and was only discontinued in 2007. The underside of the AMD version of the 80286, which had a higher clock frequency. Source: https://w.wiki/GYaQ Fig.22: an 80386DX chip. Source: https://w.wiki/GYK7 (CC BY-SA 3.0) The AM386 is a clone of the 80386. Source: https://w.wiki/GYaY Next month That’s all we have space for in this issue. We’ll pick up the rest of the Intel story in the March issue, at the start of the 1990s. That second article will bring us up to date, and then in the final instalment, we’ll look at the current state of microprocessor technology and how Intel plans to remain SC competitive in the future. siliconchip.com.au Fig.23: an exposed 80486 chip die. Source: https://w.wiki/GYKB (CC BY-SA 3.0) Australia's electronics magazine February 2026  27