Semiconductor and People (part 2.)

9 min readMar 24, 2022

“Semiconductors and People (Part. 2)”

Semiconductors are really cool. But, just like history, it commonly ignores the people who make it happen. So in this part 2 of “Semiconductors and People”, people, are exactly who we will be focusing on. In this continuation of the semiconductor story, we’re diving into people like Dov Frohman, Lee Boysel, Andy Grove (Early Intel), and Shima Masatoshi. Then, we will continue the technological timeline and begin to continue the timeline on which we continued on towards current day. This will cover process nodes, the rise and fall of lithography companies, Japanese dominance in silicon including their fall in lithography… plus much more! So buckle up, because this is going to be a fun one.

“The Inventors of the 4004”

Intel 4004 in its original ceramic casing

To really understand the silicon industry as it is today we must first dial back all the way to the 4004, the world’s first commercially available CPU. Starting off with the leader of the pack, Federico Faggin. Faggin came to Intel from SGS Fairchild in Italy, which is now STMelectronics. He was the leader of MOS research there and was called to Fairchild headquarters to continue development on Silicon Gate Technology. This proved to be influential to the 4004 as Faggin used the same methodology, combined with “Random Logic”, to design the 4004. After Fairchild, he joined Intel and began work on the 4004. The 4004 was a 4-bit single-chip microprocessor or CPU for a Busicom calculator. The 4004 was designed together with Shima Masatoshi, Ted Hoff, and Stan Mazor. Faggin had initially developed the SGT technology that made the 4004 possible, so he was the main force behind the design of the 4004 with continual aid from Masatoshi, who was here to help as he was sent by Busicom to check on the currently non-existent processor. To get the 4004 out the door these two worked overtime to make up for the lost time and completed the design in 1970. The 4004 was a marvel of a chip, aided with the fierce knowledge and experience of Faggin and the financial incentives from Busicom, the 4001, 4002, 4003, and 4004 were developed and pushed out the door for commercial use. An interesting pattern you may have noticed in early semiconductor tech is that a lot of the people who go on to design and develop chips come from Fairchild, yet the inventors of many pioneering ideas like Transistors and Photolithography came from Bell Labs, just an interesting observation.

“Creators of photolithography”

The GCA MANN 4800, one of the first auto photolithography machines

Photolithography is one of the most important technologies used in semiconductor manufacturing, it uses a PhotoResist layer to protect parts that don’t need to be etched from the etchant. It is widely thought that this idea to use photoresist came from a certain man named “Jules Andrus” and “Walter L. Bond”, since these guys were very much pioneers in this field of research I just can’t find any information on them, it is very hard to find information of Jules Andrus especially. Sadly until I find another deep source for this, I cannot verify what is on the internet. From an interview excerpt, we can observe that Andrus & Bond had a paper for using photoresists in lithography, which I could not find. These two pioneers, however, certainly did propose the idea of photoresist for lithography first. These two came from Bell Labs. Photolithography as a technology, however, moved on from its creators. The technology is heavily intertwined with the “Planar Process”, invented by Jean Hoerni in 1958. The first involvement of lithography for semiconductors as a mass volume manufacturing device came in the 50s as well, it is not certain who and what company shipped the first photolithography machine, however. One of the earlier examples I have found is the Mann 4800 by American company GCA. It is also very similar to the second Chinese lithography machine developed. The 4800, however, also was the first-ever stepper-based lithography machine, which makes it great for our description of a modern chip making machine. A stepper pretty much just moves the wafer around to put images onto it and etch it down into chips. The photolithography industry has thus then evolved in three major “eras” as I would call it, and these are:

The American era

During this time, the US had an iron grip on semiconductor machines and made most of the equipment used during the 50’s and 60’s

The Japanese era

This is the time period where Japan ruled the semiconductor industry all the way up to their economic crisis in the 90’s

The “Modern” or European era

ASML (ASMLithography at the time) PAS 5500 i-line machine advertisement

This is our era, where ASML (Formally ASMLithography, owned by Phillips) rules the market and is considered the lucky “chosen one” as it pretty much is a monopoly on the high-end EUV and high-end semiconductor manufacturing industry.

This is pretty much it for photolithography as a technology and its eras, there might be another article on each of these soon.

“Automated Design Tools and cool stuff for chip design”

Without automated design tools/philosophies like EDA,VLSI and many other intricate pieces of thinking and software, the chip industry would have never reached the scale it is now. This is because, back in the 4004 days, Faggin and his team hand drew the transistors onto paper and then traced onto RubyLith (chosen material of the time for photo masking) which was painstakingly inefficient. So during the great computer boom of the 70’s and 80’s, companies began investing in automated tools. These tools first coalesced for wiring up circuit boards and arranging wires for them, which was normal for the time as these things needed to be changed and rerouting every wire was a huge pain. This tool was then evolved into EDA, which is what we use now for trace routing on our CPU’s, saving time and effort when redesigning portions of the chip. Early tools like this just traced the result onto paper, but this quickly turned into a electronic system that drew it on monitors and was aided by the rise in CAD tools later on.

That being said, without certain philosophies and methodologies in chip design, it really is quite impossible to reach the current scale that we are at now. For me, the most important of those methodologies was VLSI, or Very Large System Integration, which was taught to a lot of students back in the day and influenced many systems like the famous Geometry Engine that was used by SGI, a prominent player in the early parallel computing or GPU industry at the time. This design philosophy enabled the creation of large-scale single-chip devices, which enabled a whole lot of opportunities for chip designers. VLSI was heralded as one of the most important inventions for semiconductors for its time, so many innovative chips came out of it. At the moment, Synopsys and Cadence are the largest EDA tool companies. Still quite a secretive and intricate industry, as with everything related to semiconductor design and manufacturing.

“Continuation of the story”

The inventors of the semiconductor transistor, using germanium at bell labs

So, the story continues. During Part 1 we finished at around 1987 when TSMC was founded, carrying on from then we can and will evaluate the chip industry towards 2010 in three ways, and they are:

Transistor Count
Process node size
Wafer size

Starting off with transistor count, we start off with the 4004 that had 2300 transistors in 1971, we then transition to the Z80 in 1976 with 8500 transistors. This period of CPU design and manufacturing was a huge leap as the 4004 was really only meant for calculators but more general-purpose CPU and micro-processors were needed as soon as it became apparent that this technology was more than it seems. Jumping forward after the Z80, the transistor count soon rose over 20 thousand and almost 30 thousand with the introduction of the Intel 8086 in 1978. This count soon rose to over a million transistors in the Intel 80486 or i486 released in 1989. This was a huge leap in performance and was widely accepted by quite a lot of PC manufacturers. This is also the time period when AMD started doing in-house x86 chip design starting off with the AMD K5. Fast forwarding just a few more years to 1999, AMD releases the K7 which has over 22 million transistors. Justs 5 years later we have the core 2 duo which sports a whopping 291 million transistors. The duo was a dual core processor, a first of its kind, and really catapulted the chip design market forward to the billion transistor mark in around 2010. This sort of complexity is super fascinating and I really believe more people should recognize the efforts and movement in the chip design industry more.

Visualization of transistor shrinkage from vacuum tubes to the modern semiconductor

And now we travel back to the 4004 in our process node analysys, it started at around 10 microns or 1/8th of a human hair. After this, however, there were huge leaps. We go to 1 micron in early 1980, which was phased out in early 1990. This marked the last micron level chips and we now go down to the nanometer (nm). We now go to around 800–600nm, the first leap to the nanometer era, this happened in around 1991–1994. We go forward just two years and arrive at 350 and 200nm nodes, this first began volume production around 1995–1996. During 1990–2000 there were multiple advances, leading to around 90nm in 2002 with Intel entering volume production then. Jumping to around 2007, TSMC begins volume production of the 40nm node, with many competitors following. We then jump to 32nm in 2009, which marks the start of a noticeable slow down of the transistor count and transistor size multiplier. Intel began doing 14nm nodes in 2014 and they still do! 14nm was one of the longest a large chip maker has sat on a singular process node. It was only until 4 years later in 2018 when Intel began doing 10nm production, which really only became implemented in its products around… now. Currently, TSMC is at 7nm and 5nm volume production with 3nm nodes coming very soon. In the next decade, we might see less than 1nm happen, as we are reaching the limits of current technology.

TSMC process node history chart, visualized.

It truly is a miracle that our species has reached this level of nanomachines and device complexity, it certainly does not end here however. Researchers have recently made a nearly 0.34nm chip with some really cool and complex technology to prevent excessive quantum tunelling between electrons. Theres going to be a lot of cool things coming on this blog as well, some more opinion articles on various things, and maybe... Perhaps… if I figure it out… a custom-designed video decode/encode core in Logisim coming soon…. If I can figure out how to do it. Chip design and manufacturing is a complex and really annoying thing so I would appreciate it if there are any people here who have industry contacts that I could ask some really beginner questions. I don't have a materials Ph.D. or some fancy electrical engineering job so I really don’t understand most of this too well. Another article will come on my writing process including my research sources and tricks. So that's it for this one!

Thank you so much for reading and have a nice day~~

Sources:

http://museum.ipsj.or.jp/en/pioneer/shima.html

https://www.mordorintelligence.com/industry-reports/semiconductor-lithography-equipment-market

http://www.vintagecalculators.com/html/busicom_141-pf_and_intel_4004.html

https://www.computerhistory.org/siliconengine/reusable-programmable-rom-introduces-iterative-design-flexibility/

https://www.computer.org/csdl/magazine/an/2018/04/08540014/17D45Wuc35a

https://intelretiree.com/wp-content/uploads/2016/03/Dov-Frohman-LAI.pdf

Physics and Technology of Semiconductor Devices, Book, Andy Grove

https://patents.google.com/patent/US2890395A/

https://ethw.org/Oral-History:Goldey,_Hittinger_and_Tanenbaum

http://www.intel4004.com/mrld.htm