Discover more from The Asianometry Newsletter
The History of the Semiconductor Photomask
From Rubylith to EUV
If you want to watch the video, it is below.
The photomask is one of three critical parts in a photolithography system.
Can you guess the other two? I will give you a minute or so to think about it.
Anyway the mask contains a master image with a chip design or part of a chip design.
As a fundamental part of the lithography puzzle, it has a fascinating history that goes all the way back to the very beginning.
So in this video, we are going to look at this essential piece of the semiconductor manufacturing process from rubylith to EUV: The photomask.
Back to Lithography
Alright time is up! Did you get it? The other two parts of the lithography process are the exposure tool and the resist.
During the lithography process, the exposure tool sends high-powered light through or off of that mask and onto the wafer.
The wafer surface is coated with a light-sensitive substance - the resist. The resist reacts to the light so that the chip pattern gets printed onto the wafer. After that, the pattern can then be etched in using further processes.
Back in the earliest days - like the 1960s or so - people used a method called contact printing to make integrated circuits. This very crude and simple method was used for making chips with feature sizes of about 200 micrometers.
Making a photomask in those days was as much kindergarten arts and crafts as it was a science.
You first start with your hand-drawn circuit design on graph paper. To enhance the design's precision, you would draw it up at much larger scale than the final actual size. Say 50, 200, or even 500 times.
Then you have to cut that design onto a special type of 2-layer photographic film called rubylith.
Rubylith has two layers: A light-blocking red film laminated to a second layer of polymer. Usually Mylar or something. You use the knife to scrape away the rubylith's softer red layer, leaving the clear polymer layer behind.
This was delicate work. And because the designs were often so large, cutters had to be mindful of their sharp toenails when they climbed all over the tables. Fresh socks were considered essential.
The completed rubylith design was often referred to as an "artwork". Rubylith itself was originally developed by the Ulano Company for screen printing and the graphic arts.
And in fact, this whole method comes out of the printing industry. The semiconductor industry adopted it for their own nefarious purposes.
So it is kind of interesting to think about the shared technical ancestors between a leading edge Apple M2 chip and your average screen printed shirt.
From Artwork to Master Mask
The next step would be to reduce the artwork down from its 500x scale. This is done in an intermediary step using special cameras. The pattern is backlit with a mercury lamp onto a photographic glass plate.
You would end up with a 10x reticle, which contains a small piece of your final mask. I love the German phrase for it though I obviously can't pronounce it: Zwischennegativ.
At this point, the reticle is still about 10x the size of the final intended size. You need to arrange many reticles in a repeating pattern onto a single 1x sized master mask.
In 1961, a company called David Mann introduced a tool called a photorepeater. It does two things: First, it reduces the reticle to its final size. Second, it prints the reticle's design onto a glass substrate slowly step by step.
This step and repeat camera as it was called is the technical precursor to today's wafer steppers.
Mask to Mask
The master mask is then used to create multiple "working masks" for the actual contact printing process.
They call it contact printing because the photomask comes directly in contact with the resist-coated wafer. You press a mask hard against the wafer in a vacuum so to get good contact and avoid any flatness errors.
Sophisticated metrology tools did not exist back then. Like at all, to be honest. People looked for errors with a steel measuring tool and a 10x magnifying eye loupe like those used by jewelers.
Obviously this manual method is not very scalable. It works fine if you are trying to get a proof of concept design out, but for high volume production you are going to want something a bit more sophisticated.
Much of the value and expense in early semiconductor manufacturing came from the actual act of making the masks rather than printing the wafers. The industry realized that they needed new technologies to improve throughput and accuracy.
One of the first improvements they added was a tool called a coordinatograph, which was borrowed from cartography and kind of looks like a light table. It mechanically adds stops in the X and Y axis so to help guide your hand in both the direction and length of the rubylith cut.
I really do love how early semiconductor people were just borrowing here and there from all sorts of different industries.
After that came artwork generators - sometimes also called pattern generators - like David Mann's Mann 3600. These use primitive computer numerical controls to improve on the original coordinatograph design and automate the tedious knife cutting work.
As these artwork generators got better at mechanical drawing, the industry was able to altogether eliminate the need to manually draw the design as well as the intermediary step of reducing that hand-drawn design from 500x to 10x scale.
Going into the 1970s, chip designers can now create the design in CAD software and use a projection camera to print out a reticle at 10x. This allowed design complexity to take another step forward.
With the contact printing method, constant contact between the working masks and the wafers cause them to quickly wear out. This not only means you have to replace the mask a lot, but you have to also increasingly watch out for print defects which ruin wafers and affect yield.
As chips got bigger and defect reduction got more important, it got to the point where you could only use a mask for 10 wafer exposures and no more. This was not economically sustainable.
To lengthen a working mask's working life, the industry swapped out the materials from soft goops to hard, more persistent metals like iron and chromium. This worked, but only to a certain extent.
So what do you do? You remove the contact variable entirely. That is what the Perkin-Elmer Corporation did in 1973 with the Micralign, the first projection aligner.
I mentioned this device earlier in another video. In that video, I mentioned how manufacturers were able to improve yield rates from 10% to 70% with the Micralign, but never exactly how.
This is how. In using a system of mirrors and lens to make it so that the mask and the wafer don't touch, the now quartz glass mask stays pristine over many more exposures.
The original Micralign worked beautifully. It can project a 2-micrometer feature design over a 4 inch wafer at a rate of about 40 wafers per hour. But its fatal flaw was that it projected one image from a mask across the whole wafer.
When feature sizes shrank into the 1 micrometer range, semiconductor manufacturers required more detail and precision than this method can provide. Things got fuzzy.
So GCA, Nikon, and Canon built on the aforementioned step-and-repeat camera's concepts to create the wafer stepper. Steppers fired small exposures across the whole wafer, trading speed and throughput for accuracy. Today’s advanced DUV lithography machines retain the concept.
The late 1970s also came with it a variety of brand new tools developed to protect, inspect, and repair the mask during the wafer exposure step.
For instance, one of KLA Instruments' breakthrough items was the KLA 100. This was a defect inspection tool that scanned and compared parts of the photomask for anomalies. It saved operators from having to stare at masks through a microscope all day.
And since the mask is now transmissive - with light passing through it - we have to worry a lot more about keeping particles away from it. So the industry introduced thin transparent membranes called pellicles.
You stretch a pellicle over a metal frame so that it keeps aberrant particles from reaching the mask. Today, pellicles are an essential part of the lithography process and not using them is sort of crazy.
E-beams and Lasers
Also by the 1970s, semiconductor designs were starting to get so complicated that traditional artwork generators were having trouble "printing" them all out.
A 64K DRAM pattern had over 100,000 components. Projecting all of them onto a 10x reticle could take up to 1-2 whole days. So how do we improve on that?
The most obvious successor was electron-beam technology, or e-beams. These are essentially small particle accelerators, sending streams of electrons at a target.
Originally developed as a microlithography method for enhancing resolution, the industry realized the technology was more suited for making masks right at 1x scale.
In 1974, Bell Labs developed MEBES or the Manufacturing Electron Beam Exposure System. Roughly speaking, MEBES fires a steady e-beam that sweeps across a table not unlike a CRT monitor.
The table moves at about 0.256 micrometers in the X and Y directions. When a pattern needs to be printed onto the reticle, the beam is allowed through. If not, then it is "blanked off".
MEBES remained the industry's mask-making technology of choice for a decade until ATEQ largely replaced it with a cheaper, more economical argon ion laser beam system.
Lasers cost less than e-beams but at the same time offered more accuracy - 0.05 micrometers - and stability.
The 1970s can be best described as a period of nonstop growth and innovation for the mask making industry. You had this flowering of brand new technologies, many of which remain in use to this day.
The 1980s on the other hand saw this Cambrian Explosion in mask making come to an end. New E-beam and laser writing technologies required very high investments while also allowing people to make more finely featured masks than what the industry actually needed at the time.
Mask making became commoditized. Some companies like Micron - seeing it as a critical differentiator - took it in-house. And with the semiconductor industry experiencing its first macroeconomic downturns, the third-party mask-making industry consolidated. Dozens exited the industry.
Companies focused not on the smallest possible resolution but rather on speeding up throughput, raising their yields, or lowering costs. This helped improve their competitiveness in the long run, but in the short term there was a lot of pain.
This dark time in the mask making industry would not end until the 1990s when the rest of the semiconductor industry finally caught up, with feature sizes shrinking below 1 micrometer.
One of the big mask makers emerging out of this era and remains so to this day is Toppan, a Japanese global printing company. Founded in 1900, Toppan first made their bones printing authenticity marks for cigarette packs and bank notes.
They then diversified into packaging, decor materials, and finally high precision electronics. They leveraged their experience making filters for the sugar refinery industry - which required making micrometer sized holes - to make photomasks.
In 2004, Toppan completed the market’s consolidation with the acquisition of Texas-based mask maker Du Pont Photomasks. An acquisition that I doubt would have been allowed today.
This year, they are spinning their photomask business off into a new independent Japanese entity - Toppan Photomasks.
Other relevant third-party companies in the industry include Compugraphics, Advanced Reproductions Corporation, and Photronics. Fabs like TSMC also offer their own mask making services.
There is so much more to talk about modern DUV photomasks. But we need to move on to the cutting edge: EUV.
The EUV Mask
EUV shrinks the laser wavelength size to 13.5 nanometers, a move that requires a re-engineering of decades of prior lithography techniques.
I have talked about EUV before in quite some detail in previous videos. Feel free to review them on your own.
Before EUV, masks and reticles were transmissive, which means that they were transparent. But with EUV, this was no longer feasible. The light wavelength is so small that even normally transparent quartz glass absorbs it. So EUV engineers decided to turn it into a mirror.
Making the EUV Mask
In order to create an EUV mask, you have to first create the mask blank - basically a purely blank EUV mirror without any chip design data on it.
Based on the SEMI P37-0613 standard, it is about 6.35 millimeters or a quarter of an inch thick and 142 millimeters by 142 millimeters large.
At its very top is the mirror layer - a multi-layer Bragg reflector with 40-50 alternating layer pairs of molybdenum and silicon. I talked about this before.
The mirror layer sits on top of what has to be a very thermally stable substrate. According to SEMI standards, the substrate can only expand or shrink some 6 to 10 parts per billion for each Kelvin. Otherwise you get print errors.
So for each millimeter of mirror substrate, that's about 0.000000006 mm of variance - 8 zeros. I think I got that right. Comment down below if I didn't.
It is presumed that the current EUV substrate is either Ultra Low Expansion glass or ULE from Corning - which was used for the Hubble Space Telescope.
Or Zerodur from the German glassmaker Schott AG, used for the Keck II telescope.
After that, a protective capping layer of Ruthenium about 2.5 nanometers thick is applied. This protects the mirror layer from damage in the later stages of the mask making process.
On the very bottom of the reticle, underneath the substrate, you have a layer of some chromium nitrate substance - not clear what - about 70-100 nanometers thick. This is for solidly clamping the reticle during the wafer exposure step.
I love the name they use for this - electrostatic chuck.
Finally, in order to turn the mask blank into something usable, you have to "pattern" it. This is done by applying an "absorber" layer of Tantalum on top of the capping layer - basically doing the same thing as the chromium patterning in DUV quartz glass masks.
All of this is hard enough as it is. But wait, there is more. ASML asked Carl Zeiss that the mask blank should have only 0.003 defects per square centimeter. Considering the reticle is 142 millimeters by 142 millimeters large, that is basically zero defects on the reticle.
Zero defectivity is one of EUV lithography’s key challenges to overcome. This requires you to protect the reticle from particles, be able to detect reticle defects, and finally be able to repair those defects when found. That means tools.
This is really important. There is no commercial point in being able to "make" an EUV mask if you do not also have the accompanying ecosystem of metrology and repair tools around them. You essentially have a multi-million dollar boondoggle.
Between the P37-1101 and P38-1102 standards, there are over 30 separate specifications that the EUV mirror has to meet. Relevant tools to measure them include the Dilatometer, Flatness Interferometer, a Transmission Electron Microscope, and a EUV reflectometer.
The big problem is that many current instruments are not sensitive enough to detect when things go wrong. The extreme nature of the technology contributes to the difficulty in inspecting it!
For instance, trying to probe the mirror layer's structures for defects is exceptionally challenging because ... well, it reflects really well. UV-based tools can only penetrate 2 to 13 of the mirror layer's 40 layers. So if the technician finds a bump on the mirror layer's surface, they cannot actually find where the offending defect is underneath the layers.
I can go on but this video has gotten quite long. I'll maybe do a video later on EUV defect science and technology later down the line. Perhaps if this video gets 5,000 likes or something.
In the span of half a century, the semiconductor industry has gone from hand-cutting arts and crafts in rubylith to ultra-sophisticated EUV mirrors made with exotic materials like Tantalum. The progress is astounding to think about.
Structurally, the mask-making industry has also changed a great deal with great consolidation into a few, very technically advanced players. It is a reflection of the semiconductor industry as a whole.
Where lithography goes from here, nobody really knows. But wherever it will go, the photomasks and reticles will go along with it.