The Scaling Cliff-Hanger

Moore’s Law has always been a cliff-hanger. Even Gordon Moore says he has never been able to see more than two generations ahead.

Moore’s Law has always been a cliff-hanger. Even Gordon Moore says he has never been able to see more than two generations ahead.

Two generations ahead from today’s 32nm/28nm generation is 16nm and immersion lithography, today’s mainstream lithography tool, may not be up to the task.

The most likely tool to perform that task is EUV, and ASML is scheduled to ship its first production EUV machine next year.

The machine is rated to achieve a throughput of 125 wafers per hour.

But the prototype EUV machine currently installed at Imec is achieving a throughput of six wafers per hour.

The target rate for the Imec machine was 60 wph. The reason why it’s performing so poorly is, says Imec, because of the dim light source.

The dimmer the light source, the longer wafers have to be exposed and, consequently, the lower the throughput.

The two providers of light sources to ASML are Cymer of San Diego and Xtreme Technologies of Aachen which is owned by the Japanese company Ushio.

Cymer’s technology is called ‘laser-produced-plasma.’

According to IEEE Spectrum, it uses a carbon dioxide laser to ‘vaporise speeding droplets of molten tin. As the resulting plasma cools, excited tin electrons relax back to lower-energy orbits, emitting EUV light that is steered through the machine and eventually used to cast patterns on a wafer’.

Cymer is looking to boost the power to ‘by shooting another laser at the tin droplet before the main laser hits it’.

This ‘pre-pulse’ will heat the tin, causing it to expand and allowing the second laser pulse to penetrate further into the droplet, creating more plasma.

Cymer says this technique will increase the power which it puts into a EUV’s optical system from 11W to 80W by the end of this year boosting throughput nearer to 60 wph.

Xtreme says it expects to be able, by the end of this year, to boost the power which it puts into EUV machines optical systems to 100W – and that this will deliver sufficient brightness to permit 60 wafer-per-hour processing.

But immersion lithography is currently delivering about 140 wph and the semiconductor manufacturers won’t want to settle for anything much slower.

Time is getting short, and he difference between 6 wph and 120 wph is huge.

If Moore’s Law scaling is to suffer a blip, then it will become all the more important to find another way to reduce die cost.

And the most likely way to do that is 450mm.



  1. You’re absolutely right, Ian, and my understanding is essentially the same as yours i.e. that the current argy-bargy is over who pays for what.

  2. Germany might have paid for the fab and process development for 300mm but I don’t think they paid the development costs of the equipment, the suppliers lost out big-time for many years which is why they don’t want to cough up again (and even more) for 450mm.
    Intel, TSMC and Samsung are quite happy to build and pay for the fabs, but not to advance the billions it will cost the equipment suppliers to develop the equipment to put in them — they want the suppliers to pay for this in the hope that they’ll get their money back, which depends on lots of fabs switching to 450mm much quicker than they did to 300mm…

  3. Absolutely true, Ian, but if someone else pays for the 450mm development, i.e. someone outside the industry (as Germany did for 300mm), then it would be a useful cost-reduction measure. And if scaling stops, then it’s the only cost reduction measure around.

  4. All true, but the time for each step in wafer size has been going up rapidly (4″==>6″, 6″==>8″, 8″==>12″) as the cost of developing the equipment rises almost exponentially, even though the total value of wafer business has also been rising.
    If the driver to 450mm is cost then the issue is whether the die cost savings in production will ever pay for the development cost, especially if this depends on the future value of wafer business continuing to rise year-on-year forever — which looks very like the house price situation in the UK to me (invest in houses, house prices will *always* keep going up faster than inflation and incomes…)

  5. You’re absolutely right of course, Ian, but if the industry had balked at the difficulties of past wafer size transitions, it would still be on two inch wafers, and that wouldn’t do at all.

  6. Die cost will only be lower for 450mm once the immense cost of developing and installing the equipment has been paid off, which could be quite some time — and even more so for EUV which will have even more of a throughput problem with the bigger wafers.

Leave a Reply

Your email address will not be published. Required fields are marked *