EUV Might Really Happen

I have been a skeptic about whether EUV was going to work. Just in case you have no idea what I'm talking about, EUV stands for extreme ultra-violet. For what seems like forever, we have been using 193nm wavelength light for lithography, eventually adding immersion so it often is called 193i. At 13.5nm wavelength, EUV would at least not need to be double patterned at 7nm.

At one of the short courses on 5nm design at IEDM, one of the speakers was Anthony Yen of TSMC. He actually gave a tutorial on lithography and what terms like "numerical aperture" really mean and how you can prove the resolution formula. But for me the most interesting part was at the end when he summarized TSMC's current experience with EUV. I think it was a lightly updated version of his talk at SEMICON Taiwan a couple of months ago. He started with a slide from 1999 looking at future possibilities for future lithography. The only one left alive is EUV—the entire industry is all-in as they say in TV poker.

EUV has four big outstanding problems, and Anthony went over them. They are:

• Source power (tied up closely with throughput)
• Defect free masks
• Pellicle or rather lack of it
• Availability (tied up closely with cost)

The light source for EUV is produced by what seems, when you first hear about it, to have been designed by Rube Goldberg. Droplets of tin, 50,000 of them per second, are generated and fall. A laser is used to hit each one twice, once to shape the droplet of tin so that the next phase will be more effective. The second is a powerful blast of the laser that actually produces the light. Oh, and I forgot to mention, EUV is absorbed by almost everything so the entire system has to be in a full vacuum.

The problem with the light source has been to get the power up enough. It is generally thought that it needs to be 250W for it to be acceptable in volume production. Things have been going well in this area as the chart below shows. The red triangles are results TSMC have achieved with the EUV stepper they have in Taiwan. The blue are results that ASML in the Netherlands have achieved with their generator.

I said EUV is absorbed by everything. That includes lenses. So EUV has to have reflective optics. But EUV would be absorbed by the kind of mirror in your bathroom. Instead they have to be build up with alternating layers of silicon and molybdemum and rely on a phenomenon called Bragg reflection. Only about 60% of the light is reflected, so heat is an issue (the first mirror, for example, is absorbing about 40% of 250W). Also, since there are six mirrors (including the mask) only a few percent of the generated light actually hits the photoresist.

The mask is a similar type of mirror, but patterned, obviously (see the diagram to the right). One issue with masks is that there is not yet a means to make them defect free. The number of defects has steadily fallen, and is now around 20. If the defects can all be identified and the mask-making equipment can position accurately enough, then it is hoped that it will be possible to hide the defects behind the mask in the areas where reflection is not required.

But the biggest problem with the mask system at present is the lack of a pellicle. On a normal 193i mask, there is a cover on the mask so that if any contamination lands on it then it is not in the focal plane. EUV masks need the same thing otherwise a contaminant would sit on the mask wrecking wafer after wafer until the mask was cleaned. Worse, it might not show up when the wafers were inspected so a lot more useless processing would also be wasted. But finding a material for the pellicle is hard. Remember, EUV is absorbed by almost everything. The only material so far that has been considered a possibility is a polysilicon membrane. TSMC have made membranes large enough to cover the whole wafer. But testing them has been disappointing.

The problem is that with the membrane 55nm thick, then only 85% of the light gets through which is considered unacceptable. They will need to make the pellicle thinner but at some point it must start to get too close to the focal plane to do its job.

But when it all works, it works well. This is an especially difficult sort of pattern for lithography. On the left is the design double patterned using 193i. On the right is EUV. The little tiny fingers are the stress point where you can see the double patterning just doesn't have powerful enough OPC to do the job.

So what about availability? That has always been the big worry, that when it works it works well, but the machine is very unreliable. TSMC has been running wafers and processed 15,000 wafers in four weeks. Tool availability was just over 70%.

So are TSMC going to introduce EUV at 7nm? Anthony said he wasn't going to commit to that. "When it is ready we will introduce it."