Guest author Ken Shirriff is a Silicon Valley-based figurer enthusiast who enjoys reverse-engineering former chips and restoring classic equipment such equally the Xerox Alto. Shirriff wrote the Arduino IRremote library for infrared remotes, attempted Bitcoin mining on a 55 year sometime IBM 1401 dial bill of fare mainframe, and got six symbols added to Unicode including the Bitcoin symbol (฿). Y'all tin can follow his work at Righto.com and @kenshirriff.

Intel'south groundbreaking 8008 microprocessor was starting time produced 45 years ago.1 This bit, Intel's first 8-bit microprocessor, is the ancestor of the x86 processor family that you may be using right now. I couldn't detect expert die photos of the 8008, and so I opened one upwards and took some detailed photographs. These new die photos are in this article, along with a give-and-take of the 8008's internal pattern.


Die photograph of the 8008 microprocessor

The photo in a higher place shows the tiny silicon die inside the 8008 package (click the image for a college resolution photograph). You lot tin can barely encounter the wires and transistors that brand up the flake. The squares around the outside are the 18 pads that are connected to the external pins by tiny bond wires. You tin can see the text "8008" on the right border of the chip and "© Intel 1971" on the lower edge. The initials HF appear on the tiptop right for Hal Feeney, who did the scrap'southward logic design and physical layout. (Other primal designers of the 8008 were Ted Hoff, Stan Mazor, and Federico Faggin.)

Inside the chip

The diagram below highlights some of the major functional blocks of the chip. On the left is the 8-fleck Arithmetic/Logic Unit (ALU), which performs the actual data computations.3 The ALU uses two temporary registers to hold its input values. These registers have upwardly significant area on the fleck, not considering they are complex, only because they need large transistors to bulldoze signals through the ALU circuitry.


Die of the 8008 microprocessor showing major components.

Below the registers is the conduct expect ahead circuitry. For addition and subtraction, this circuit computes all eight comport values in parallel to improve performance.two Since the low-order conduct depends on just the depression-order bits, while the higher-social club carries depend on multiple $.25, the excursion block has a triangular shape.

The triangular layout of the ALU is unusual. Most processors stack the circuitry for each bit into a regular rectangle (a bit-slice layout). The 8008, however, has eight blocks (i for each bit) arranged haphazardly to fit around the space left past the triangular acquit generator. The ALU supports eight simple operations.three

In the middle of the chip is the instruction register and the teaching decoding logic that determines the significant of each eight-flake automobile pedagogy. Decoding is done with a Programmable Logic Assortment (PLA), an arrangement of gates that matches bit patterns and generates the appropriate control signals for the residue of the chip. On the right are the storage blocks. The 8008's seven registers are in the upper right. In the lower right is the address stack, which consists of eight 14-bit address words. Unlike most processors, the 8008's phone call stack is stored on the chip instead of in memory. The program counter is just one of these addresses, making subroutine calls and returns very uncomplicated. The 8008 uses dynamic memory for this storage

The concrete construction of the flake is very close to the block diagram in the 8008 User's Manual (below), with blocks located on the chip in virtually the aforementioned positions as in the block diagram.


Block diagram of the 8008 microprocessor, from the User's Manual.

The structure of the chip

What does the die photo testify? For our purposes, the scrap can be thought of every bit three layers. The diagram below shows a closeup of the chip, pointing out these layers. The topmost layer is the metal wiring. It is the most visible feature, and looks metallic (not surprisingly). In the particular beneath, these wires are by and large horizontal. The polysilicon layer is beneath the metal and appears orange under the microscope.


A closeup of the 8008 die, showing the metallic layer, the polysilicon, and the doped silicon.

The foundation of the bit is the silicon wafer, which appears purplish-gray in the photo. Pure silicon is effectively an insulator. Regions of it are "doped" with impurities to create semiconducting silicon. Being on the lesser, the silicon layer is hard to distinguish, but you tin run across black lines along the border between doped silicon and undoped silicon. A few vertical silicon "wires" are visible in the photo.4

Transistors are the fundamental component of the chip, and a transistor is formed where a polysilicon wire crosses doped silicon. In the photograph, the polysilicon appears as a brighter orange where it forms a transistor.

Why an eighteen pin chip?

I inconvenient feature of the 8008 is information technology only has 18 pins, which makes the chip slower and much more difficult to employ. The 8008 uses fourteen address bits and 8 data bits so with 18 pins there aren't enough pins for each point. Instead, the flake has 8 data pins that are reused in three cycles to transmit the low address bits, high accost bits, and data bits. A reckoner using the 8008 requires many support chips to interact with this inconvenient bus architecture.v

There was no good reason to force the chip into 18 pins. Packages with forty or 48 pins were common with other manufacturers, just sixteen pins was "a faith at Intel".6 Only with slap-up reluctance did they move to 18 pins. By the time the 8080 processor came out a few years later, Intel had come up to terms with 40-pin chips. The 8080 was much more popular, in function because it had a simpler bus design permitted by the 40-pin package.

Power and data paths in the chip

The information charabanc provides data flow through the chip. The diagram below shows the viii-bit data bus of the 8008 with rainbow colors for the eight data lines. The data bus connects to the 8 information pins along the exterior of the upper half of the flake. The bus runs between the ALU on the left, the instruction register (upper center), and the registers and stack on the right. The omnibus is separate on the left with half forth each side of the ALU.


Die photograph of the 8008 microprocessor. The power motorbus is shown in ruby-red and blueish. The data bus is shown with 8 rainbow colors.

The reddish and blue lines show power routing. Ability routing is an under-appreciated aspect of microprocessors. Ability is routed in the metal layer due to its low resistance. Simply since in that location is only one metal layer in early microprocessors, power distribution must be carefully planned then the paths don't cross.vii The diagram in a higher place shows Vcc lines in blue and Vdd lines in red. Ability is supplied through the Vcc pin on the left and the Vdd pivot on the right, then branches out into thin, interlocking wires that supply all parts of the chip.

The annals file

To show what the fleck looks like in item, I've zoomed in on the 8008'due south register file in the photograph below. The register file consists of an eight past seven grid of dynamic RAM (DRAM) storage cells, each using iii transistors to hold one bit.eight (You lot can encounter the transistors as the pocket-size rectangles where the orange polysilicon takes on a slightly more vivid color.) Each row is one of the 8008's seven 8-fleck registers (A, B, C, D, E, H, 50). On the left, you can see seven pairs of horizontal wires: the read select and write select lines for each register. At the top, you can see eight vertical wires to read or write the contents of each bit, along with 5 thicker wires to supply Vcc. Using DRAM for registers (rather than the more common static latches) is an interesting pick. Since Intel was primary a memory company at the fourth dimension, I expect they chose DRAM due to their expertise in the surface area.


The register file in the 8008. The fleck has seven eight-bit registers: A, B, C, D, East, H, L

How PMOS works

The 8008 uses PMOS transistors. To simplify slightly, you can think of a PMOS transistor as a switch betwixt two silicon wires, controlled past a gate input (of polysilicon). The switch closes when its gate input is low and information technology can pull its output high. If you lot're familiar with the NMOS transistors used in microprocessors like the 6502, PMOS may be a bit confusing because everything is backwards.

A elementary PMOS NAND gate can be constructed as shown below. When both inputs are high, the transistors are off and the resistor pulls the output low. When any input is low, the transistor will bear, connecting the output to +v. Thus, the circuit implements a NAND gate. For compatibility with v-volt TTL circuits, the PMOS gate (and thus the 8008) is powered with unusual voltages: -9V and +5V.


A NAND gate implemented with PMOS logic.

For technical reasons, the resistor is really implemented with a transistor. The diagram beneath shows how the transistor is wired to human activity as a pull-down resistor. The detail on the right shows how this circuit appears on the chip. The -9V metal wire is at the top, the transistor is in the eye, and the output is the silicon wire at the bottom.


In PMOS, a pull-downward resistor (left) is implemented with a transistor (center). The photo on the right shows an actual pull-down in the 8008 microprocessor.

History of the 8008

The 8008's complicated story starts with the Datapoint 2200, a popular computer introduced in 1970 as a programmable final. (Some people consider the Datapoint 2200 to be the beginning personal computer.) Rather than using a microprocessor, the Datapoint 2200 contained a board-sized CPU build from individual TTL fries. (This was the standard way to build a CPU in the minicomputer era.) Datapoint and Intel decided that information technology would be possible to replace this lath with a single MOS chip, and Intel started the 8008 project to build this flake. A scrap later, Texas Instruments also agreed to build a single-chip processor for Datapoint. Both chips were designed to be uniform with the Datapoint 2200's eight-bit didactics set and compages.


The 8008 processor was first described publicly in "Electronic Design", Oct 25, 1970. Although Intel claimed the chip would be delivered in January 1971, actual delivery was more than a yr later in April, 1972.

Around March 1971, Texas Instruments completed their processor fleck, calling it the TMC 1795. After delaying the projection, Intel finished the 8008 chip subsequently, around the end of 1971. For a variety of reasons, Datapoint rejected both microprocessors and built a faster CPU based on newer TTL fries including the 74181 ALU chip. TI tried unsuccessfully to marketplace the TMC 1795 processor to companies such every bit Ford, but ended up abandoning the processor, focusing on highly-assisting calculator fries instead. Intel, on the other hand, marketed the 8008 every bit a general-purpose microprocessor, which eventually led to the x86 architecture you're probably using right now. Although TI was commencement with the viii-fleck processor, it was Intel who made their chip a success, creating the microprocessor industry.


A family unit tree of the 8008 and some related processors. Blackness arrows indicate backwards compatibility. Light arrows indicate significant architecture changes.

The diagram to a higher place summarizes the "family tree" of the 8008 and some related processors.ten The Datapoint 2200's architecture was used in the TMC 1795, the Intel 8008, and the adjacent version Datapoint 220011. Thus, four entirely different processors were built using the Datapoint 2200's instruction set and architecture. The Intel 8080 processor was a much-improved version of the 8008. It significantly extended the 8008's educational activity ready and reordered the machine code instructions for efficiency. The 8080 was used in groundbreaking early microcomputers such every bit the Altair and the Imsai. Afterwards working on the 4004 and 8080, designers Federico Faggin and Masatoshi Shima left Intel to build the Zilog Z-fourscore microprocessor, which improved on the 8080 and became very popular.

The jump to the xvi-bit 8086 processor was much less evolutionary. About 8080 assembly code could be converted to run on the 8086, but not trivially, as the teaching fix and architecture were radically changed. Nonetheless, some characteristics of the Datapoint 2200 still exist in today's x86 processors. For instance, the Datapoint 2200 had a series processor, processing bytes 1 bit at a time. Since the lowest bit needs to be candy first, the Datapoint 2200 was lilliputian-endian. For compatibility, the 8008 was little-endian, and this is still the case in Intel's processors. Another characteristic of the Datapoint 2200 was the parity flag, since parity calculation was important for a terminal'south communication. The parity flag has continued to the x86 compages.

The 8008 is architecturally unrelated to Intel's 4-bit 4004 processor12. The 8008 is not an 8-scrap version of the 4-fleck 4004 in any way. The similar names are purely a marketing invention; during its pattern phase the 8008 had the unexciting name "1201".

If you desire more early microprocessor history, I wrote a detailed commodity for the IEEE Spectrum. I also wrote a post nigh TI's TMC 1795.

How the 8008 fits into the history of semiconductor technology

The 4004 and 8008 both used silicon-gate enhancement-mode PMOS, a semiconductor technology that was only used briefly. This puts the chips at an interesting betoken in chip fabrication engineering science.

The 8008 (and modern processors) uses MOS transistors. These transistors had a long path to acceptance, beingness slower and less reliable than the bipolar transistors used in most computers of the 1960s. By the late 1960s, MOS integrated circuits were condign more common; the standard technology was PMOS transistors with metallic gates. The gates of the transistor consisted of metallic, which was also used to connect components of the chip. Chips substantially had two layers of functionality: the silicon itself, and the metallic wiring on top. This technology was used in many Texas Instruments calculator chips, equally well as the TMC 1795 chip (the chip that had the same teaching set every bit the 8008).

A key innovation that made the 8008 practical was the self-aligned gate—a transistor using a gate of polysilicon rather than metal. Although this technology was invented by Fairchild and Bell Labs, it was Intel that pushed the technology ahead. Polysilicon gate transistors had much better functioning than metal gate (for complex semiconductor reasons). In addition, adding a polysilicon layer fabricated routing of signals in the chip much easier, making the fries denser. The diagram beneath shows the do good of self-aligned gates: the metallic-gate TMC 1795 is bigger than the 4004 and 8008 fries combined.


Intel's 4004 and 8008 processors are much denser than Texas Instruments' TMC 1795 fleck, largely due to their use of self-aligned gates.

Shortly afterwards, semiconductor engineering science improved again with the use of NMOS transistors instead of PMOS transistors. Although PMOS transistors were easier to manufacture initially, NMOS transistors are faster, then once NMOS could exist fabricated reliably, they were a clear win. NMOS led to more powerful chips such as the Intel 8080 and the Motorola 6800 (both 1974). Another engineering science improvement of this time was ion-implantation to change the characteristics of transistors. This allowed the cosmos of "depletion-mode" transistors for use as pull-upwards resistors. These transistors improved chip performance and reduced ability consumption. They too allowed the cosmos of chips that ran on standard 5-volt supplies.13 The combination of NMOS transistors and depletion-mode pull-ups was used for nearly of the microprocessors of the late 1970s and early 1980s, such as the 6502 (1975), Z-80 (1976), 68000 (1979), and Intel fries from the 8085 (1976) to the 80286 (1982).

In the mid 1980s, CMOS took over, using NMOS and PMOS transistors together to dramatically reduce power consumption, with fries such as the 80386 (1986), 68020 (1984) and ARM1 (1985). At present virtually all chips are CMOS.fourteen

As you tin can run into, the 1970s were a time of large changes in semiconductor scrap technology. The 4004 and 8008 were created when the technological adequacy intersected with the right market.

How to accept die photos

In this section, I explain how I got the photos of the 8008 dice. The kickoff footstep is to open up the bit parcel to expose the die. Virtually chips come up in epoxy packages, which tin exist dissolved with unsafe acids.


The 8008 microprocessor in a ceramic packet

Since I would rather avoid boiling nitric acrid, I took a simpler approach. The 8008 is as well available in a ceramic packet (higher up), which I got on eBay. Tapping the chip along the seam with a chisel pops the two ceramic layers apart. The photo beneath shows the lower half of the ceramic parcel, with the die exposed. About of the metallic pins have been removed, but their positions in the package are visible. To the right of the dice is a pocket-size square; this connects ground (Vcc) to the substrate. A couple of the tiny bond wires are still visible, connected to the die.


Inside the package of the 8008 microprocessor, the silicon die is visible.

In one case the die is exposed, a microscope can exist used to take photographs. A standard microscope shines the calorie-free from beneath, which doesn't work well for die photographs. Instead, I used a metallurgical microscope, which shines the lite from above to illuminate the flake.

I took 48 photographs through the microscope and then used the Hugin stitching software to combine them into one loftier-resolution image (details). Finally, I adjusted the prototype dissimilarity to make the fleck's structures more than visible. The original image (which is approximately what yous see through the microscope) is below for comparing.


Dice photo of the 8008 microprocessor

Decision

I took detailed dice photos of the 8008 that reveal the circuitry information technology used. While the 8008 wasn't the get-go microprocessor or even the first 8-bit microprocessor, it was truly revolutionary, triggering the microprocessor revolution and leading to the x86 compages that dominates personal computers today. In future posts, I plan to explain the 8008'southward circuits in particular to provide a glimpse into the roots of todays computers.

I announce my latest blog posts on Twitter, and so follow me at kenshirriff. Or yous can employ the RSS feed.

Notes and references

i. According to the oral history of the 8008, photos of the 8008 were obtained in October / Nov 1971 (page half dozen). Chip designer Federico Faggin mentions that toward the cease of 1971, "everything was working except for a few errors." Faggin then debugged a trouble with the dynamic memory losing data, making it ready for production (page ix). ↩

2. Using the carry expect alee circuit avoids the filibuster from a standard ripple-conduct adder, where the carries propagate through the sum. ↩

3. The 8008's ALU supports eight operations: add, subtract, add with acquit, subtract with carry, AND, OR, XOR, and compare. Information technology too implements left and right shift and rotate operations. The 8008 also has increase and decrement instructions, extending the Datapoint 2200's education set. ↩

4. Because silicon has higher resistance than polysilicon, most chips utilize the polysilicon and metal layers for wiring, not the silicon layer. The 4004 and 8008 chips are unusual in that they prefer to use the silicon layer for wiring rather than polysilicon. I await this was due to the contempo introduction of polysilicon: before polysilicon, routing needed to be done in the silicon layer and maybe the chip designers were sticking with the older layout techniques. ↩

5. The 8008 required 20 support chips according to flake architect Federico Faggin. In dissimilarity, the 4004 and before MOS computers such as the Four Stage and CADC were designed with a small number of MOS chips that worked together without extra "glue chips". In this sense, the 8008 was a step backwards architecturally, saying "here's the CPU, yous figure out how to make a computer out of it." ↩

6. For details on Intel'south insistence on xvi pins, see Oral History of Federico Faggin, page 55-56. It was simply when the 1103 retention bit required 18 pins that Intel reluctantly moved beyond 16 pins. And that was treated by Intel similar "the sky had dropped from heaven," resulting in "and so many long faces". ↩

7. If two metal lines demand to cantankerous, one of them can be routed under the other by using the polysilicon layer. To be low resistance, this cross-under must be relatively wide, and so cross-unders are avoided if possible. ↩

viii. The 8008 registers apply the "3T1C" cell: three transistors and 1 capacitor (details). The excursion doesn't physically contain a separate capacitor, merely uses the gate capacitance of the transistor. One unusual feature of the 8008 cell is it uses one wire for both reading and writing the scrap, while the typical 3T cell has divide wires for reading and writing. The 4004 had divide wires, but the design inverse slightly in the 8008. ↩

9. Pull-upward resistors in later chips such as the 6502 were implemented using depletion-fashion NMOS transistors. These yielded more faster, more efficient logic. They were also wired differently, with the gate connected to the output rather than the power runway. ↩

ten. The 8008 architecture and the development of Intel's microprocessors are discussed in detail in Intel Microprocessors: 8008 to 8086. ↩

eleven. The second version of the Datapoint 2200 had a totally new implementation of the processor, still built from TTL chips. While the first version had a serial ALU (processing ane bit at a fourth dimension), the second version operated in parallel using 74181 ALU chips. As a upshot, the 2d version was much faster. ↩

12. The extensive 4004 Ceremony Project has contrary-engineered the 4004 processor. The 4004 schematic is here. ↩

13. The Motorola 6800 microprocessor originally used enhancement-mode transistors. To operate off a unmarried +5V supply, it had a voltage-doubler excursion on the chip. ↩

14. Interestingly, in 2007 Intel started using metal gates over again in society to scale transistors further (details). In a way, semiconductor technology has gone full circumvolve, back to metal gates, although now unusual metals such as hafnium are used. ↩