Jerónimo López

@jerolba.bsky.social

Javero picateclas en Clarity AI y Jerocleta Manager

In 1980, Intel announced the 8087 Math Coprocessor, a chip that made floating-point 100 times faster. I opened up the chip, took photos of the silicon structures, and analyzed its circuitry. It's a very complex chip for its time. Let's take a look inside...

A photo of the 8087 die under a microscope. The die is rectangular, with complex patterns in purplish-brown. The patterns consist of rectangular regions, striped regions in the bottom half of the chip, and other more irregular regions.

At the right, two regions are highlighted in red: the registers and the stack control circuitry.

Around the edges of the die, you can see the hair-thin bond wires that connect the chip to its 40 external pins. The complex patterns on the die are formed by its metal wiring, as well as the polysilicon and silicon underneath. The bottom half of the chip is the "datapath", the circuitry that performs calculations on 80-bit floating point values. At the left of the datapath, a constant ROM holds important constants such as π. At the right are the eight registers that form the stack, along with the stack control circuitry. The chip's instructions are defined by the large rectangular microcode ROM in the middle.

Intel's 386 processor (1985) was critical to the success of Intel. With 285,000 transistors, it was too much for Intel's design process and the schedule started slipping. Intel pivoted to "standard cells", an automated technique for chip layout to get back on track. Let's look closer...

A die photo of the 386 processor. It is a square with complicated patterns on top. Under the microscope, the circuits appear in dark purple. Parts of the chip have been marked with boxes: these are standard cell circuits and have a distinctive striped appearance.

Baterías: "Cuanto más baja el precio, más se acelera su adopción, y cuanto más se acelera su adopción, más baja el precio." Y en diciembre empieza la producción en masa de baterías de sodio, cuyo coste puede llegar a ser un 90% menor que las de litio actuales. Es hora de electrificarlo todo.

Bild

In the 1960s, this absurdly large camera was used to make masks for integrated circuits. The layers of the integrated circuit were drawn at large scale and then optically shrunk to make the glass masks that were used in manufacturing.

Black and white photo of a giant bellows camera about 8 feet tall. A man is standing in front of the lens which is the size of his torso. Caption: "A copy camera of the type useful in making high-precision photographic masks."

This photo of magnetic core memory popped up, taken by Ansel Adams. I investigated a bit: this memory holds 4000 bits, each stored in a tiny magnetized ferrite ring. It may be from an IBM 705 vacuum-tube business computer. 1/N

A photo with the caption "'Hands weaving magnetic-core memory, IBM, Poughkeepsie, New York', 1956. Photograph by Ansel Adams. This photograph was made on a commercial assignment for IBM."

This black-and-white photo shows a core memory plane being assembled. It consists of an 80 by 50 grid of wires in a frame with tiny toroids on the wire intersections. Each wire is attached to metal terminals on the frame. Someone is using tweezers to thread the tiny cores onto the wires. The plane is about 4/5 completed. One hand is below the plane, palm-up, visible through the wires. The other hand (with nail polish) is above the plane, holding tweezers.

How thin is spider silk compared to the wiring on a chip? Here's a close-up of a spiderweb strand diagonally across an Intel 386 processor. The spiderweb is 4 µm thick, just a bit thinner than this 1985 chip's 5 µm metal wiring. Wiring in modern chips is 100 times thinner.

A close-up of the die of the Intel 386 processor showing the metal layers. A strand of a spider web runs diagonally across the image. The processor image consists of yellowish metal wiring running horizontally and vertically, with more complex patterns underneath. Black circles indicate connections between the two metal layers. The spider web is about the same thickness as the metal wiring. The silicon circuitry underneath is not visible, but its feature size is 1.5 µm, considerably thinner than the spider web.

Anyone use Intel's 386 processor from 1985, the first 32-bit x86? To improve performance, it preloaded instructions into a 16-byte prefetch queue (highlighted in red). Why does such a small queue take up so much of the chip? Let's look under the microscope at its complicated circuitry... 1/N

A die photo of the Intel 386 processor. It looks a bit like an aerial photo of a weirdly-zoned city, with rectangular blocks of various textures in a complicated arrangement. The functional units of the chip are labeled. In particular, the Prefetch Unit is near the upper left corner. A large red rectangle is labeled Prefetch Queue.