Envisioning rotating spherical droplets of tin

The tin droplets are shot out of a tiny nozzle, squeezed out by some sort of piezo vibrator. I don't think they rotate much but they sure wobble.

One big issue for tin droplet detection is that the sphere isn't a nice round sphere, but has multiple vibration modes. There is a mess of higher frequency ripples sloshing all over the liquid surface scattering light everywhere. The detector output looks like a lot of noise.

There are several drops in mid-air at once and when the giant CO2 laser hits one, the shock wave whacks all the incoming droplets and makes things worse. They call it fratricide.

The detector was all analog and had to be done fast. It couldn't be the classic constant-fraction discriminator that physicists love so much.

It didn't help that Certain Parties vetoed some of our better ideas.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

I recall one such instance

RS, the co-founder of Cymer, reprimanded me for giving him your book. He said he worked his way through the thing cover to cover and didn't get anything else done for three days.

I recall that they were making a few watts of EUV in those days. I think they are pushing a kilowatt now.

I've never understood how they can do nanometer lithography with what is basically a fuzzy-ball incoherent light source.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

If the droplets had been injected into a strong magnetic field, that would have acted to damp the ripples. Make it a rotating magnetic field and you'd have spinning droplets with fixed axis of rotation.

Why not? They can be pretty fast, but they do depend on delaying a portion of the pulse.

Sloman, A.W. and Swords, M.D. "A fast and economical gated discriminator", Journal of Physics E: Scientific Instruments, 11,

delayed a portion of the pulse by 1.6nsec to compare it with a portion of the undelayed pulse. That delay does depend on the shape of the pulse you are looking at. Ours was about 2.5nsec wide.

A slow reader. It's a fascinating book and quite bulky, but it didn't take me three days to proof-read it.

That's what apertures are for. If the fuzzy ball is far enough away, and bright enough, it can become a pretty narrow angle source.

Too noisy. A proper droplet detector lowpass filters the worst noise and then finds the pulse centroid, over a serous range of pulse widths and amplitudes.

It's really a statistical game.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

You really do need to define what you mean by "too noisy".

You seem to be claiming that the signal you are looking at has range of pulse widths, and you have to hit your droplet with your laser shortly after the signal has peaked, independent of the shape of the rising edge.

Presumably you have one or more low powered light source illuminating the volume where the tin droplet will appear, and several photodetectors that can detect the light reflected off the droplet.

If the droplet is vibrating - as you say it is - each detector will see an occasional photon as the droplet grows, and stop seeing them as the droplet flies beyond the illuminated space.

Summing the output of several detectors should give you a tolerably well-behaved signal.

The droplets aren't moving all that fast, so we aren't talking about nanosecond signal processing here.

Seems obvious. Any amount of noise it locating the droplet reduces wafer fab rate, and costs money. A CFD assumes that every pulse is identical in shape and is noise-free. Nice theory.

We want to hit the droplet dead center, regrdless of the optical uncertanties.

The droplet passed through a sheet laser and reflected back into a single photodiode. That's what we had to work with.

One microsecond is a tolerable error.

Of course ASML has moved on in 20+ years. I think (from public sources) that they are now actively steering the droplets into the target zone and surely have better optics.

The process is interesting if horrendous. There's stuff online and no doubt patents. Turns out that wafer throughput is worth a lot.

Many people are spending big bucks to supercede the tin droplet lithography thing... including just giving up on Moore's Law.

Personally, I don't much need bigger or faster CPUs or DRAM and I don't need 10 terabytes of solid-state drives in my PCs. Maybe digital semiconductors are like dishtowels and hammers now, good enough.

I would like a sane and stable operating system.

Analog chips and power semiconductors have a way to go still, but they don't need nanometer features.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

But small.

Joe

.

Joe

Am 31.05.26 um 00:19 schrieb john larkin:

I dont't think that wobbling or rotating of the drops plays any role. When the tin really is used, it is a 200000 °C hot plasma; 40 times hotter than the surface of the sun. That surely comes with a volume increase, and that at 50000 times a second.

There is a large collector mirror, they seem to take all the light that they can get. The precision optics has a numerical aperture of

The mirrors have 100 layers of coating, each only a few atoms thick. If the mirrors were magnified to the size of Germany, the worst surface errors were < 100 um.

German: <

English: <

Light source and EUV optics by Zeiss SMT in Oberkochen, .de, lasers by Trumpf near Stuttgart, .de

Sytem by ASML + a large network.

Gerhard

You've left out most of your reasoning there. A droplet is either there

- signal and noise - or not there - no signal and much less noise.

If you get too picky you will miss droplets, get less EUV and process the wafer more slowly.

Constant fraction discriminators do assume a constant pulse shape. There's no assumption that there's no noise - the real need for constant fraction discriminators comes from the statistical noise on the output of photomultiplier tubes where the electron ejected from the photocathode can produce two, three or four (if you get lucky) secondary electrons at the first dynode. The RCA 8885 did better, but their first dynode surface was hard to make and easy to poison by running the last dynodes too warm.

Somewhat ambitious. The optical uncertainies mean that you can't know where the dead center is.

There's no such thing a sheet laser. You might manipulate a laser beam into flat fan shape with sufficiently fancy optics. Is the sheet thinner than the droplet diameter?

That makes it a pretty slow system.

Wafer throughput is what makes the money. Fabs exist to churn out wafers which the semiconductor industry dices, packages and sells.

The fact that you can't see a way in which nanometer features might be useful reflects your unfortunate ignorance. People do weird stuff at the individual device level.