Back at the end of August, we spoke briefly about how silicon carbide (SiC) wafers are processed. My point there wasn't how chemically pure SiC wafers are turned into transistors, but how SiC wafers were handled. The main point of that article was how a company had developed a new way of handling the wafers; instead of slicing them apart with diamond wire saws, they used a laser to break chemical bonds a few molecules below the surface of crystal, letting the slice almost slide off with no effort. The resulting wafer still needed to be polished, as was the case in the old method, but the new approach yielded massive improvements in efficiency. They
produced 50% more wafers through reduced material losses and produced wafers in 16% of the time of the old method.
First a note. Some people may have very minimal knowledge of electronics. You may want to see my "The Least You Should Know" series, in particular Electronics 1 and Electronics 2.
The question came up, though, "just how do they make those complex integrated circuits?", and it's a deep question. The short answer is by photographic processes, but that doesn't really tell you much. Especially in the modern geometries we speak about - 14nm and smaller. Here, features are the size of a group of atoms; perhaps 50 or 60 atoms. Photographically, you can reduce an image, but can you reduce it that far? It doesn't take place with one photographic process, there are multiple steps and they require the "masks" to align to very high precision. Light can't be used to shine through photographic masks (just dark and light areas) because light waves are too big. Generations of transistors ago, designers compensated photo masks for how the diffraction of light would degrade the features they were trying to create. Now they've gone to shorter and shorter wavelengths, extreme ultraviolet light, to keep the wave nature of light from destroying their work.
Wait. Masks? Photographic processes? Extreme ultraviolet light? How incredibly exotic. How did transistors ever get invented and made? A few years ago, I discovered this 2010 video where self-taught engineer Jeri Ellsworth describes making a type of FET (Field Effect Transistor) at home with specialized but homemade equipment and the right chemicals. It's 8 minutes long, but it captures the essence of the process. To begin with, the vast majority of electronics hobbyists would say it's not possible to make a transistor at home, without a semiconductor fab. But the first transistors and first integrated circuits were made before fabs existed, and while they were crude by today's standards, they worked. If one can gather the equipment and supplies required, they can make functional transistors this way. Here she demonstrates the transistors working.
A silicon wafer is first heated, and exposed to steam at high
temperatures to create an oxide layer. Then it's coated with a
photoresist that cures on the wafer, exposed to light which causes the photoresist to break down and expose some areas of the silicon oxide layer. The wafer is then
treated with Hydrofluoric Acid (nasty stuff!) to etch away the silicon
oxide, exposing bare silicon again. The bare silicon is treated with yet more chemicals, and this process repeated. The process with complex integrated circuits, which can include transistors, diodes, resistors, capacitors (and very rarely because they take so much room on the wafer - inductors), is essentially the same as what Jeri does, just repeated many more times and using some different chemicals. The manufacturing of a complex chip like a Digital Signal Processor can have over a thousand or 1500 operations from bare wafer to final product. It can require a few hundred additions of materials, and etching some or most of it away.
A couple of entertaining videos are here, only a few years old, and a considerably older video from Phillips here. While the Phillips video talks about geometries and numbers of parts that are rather dated, those processes are still used, as we also talked about.
This is a very deep subject; there are multiple textbooks available. I can only touch the surface here and give you a very superficial idea of how it's done, but it's still remarkable stuff.
This is the die of a Texas Instruments Digital Signal Processor, a TMS320C25, a "second generation DSP" - I've worked with these. DSPs are specialized microprocessors that are very good at floating point operations and especially signal processing algorithms, like a Discrete Fourier Transform. I don't know how many transistors are in this, but would guess around a billion. The low power magnification here can barely hint at the complexity that would be seen at high powers.
Thanks Graybeard! After all these years, still fascinating stuff. I started college back in the discreet component days and just when Intel got started with the 8008. There is so much that the public takes for granted.
ReplyDeleteThank you for this thrilling explanation; it is wonderful!
ReplyDelete