UV from mercury rectifiers...

If they glow, this is the place to be
User avatar
IslandPink
Amstrad Tower of Power
Posts: 10041
Joined: Tue May 29, 2007 7:01 pm
Location: Denbigh, N.Wales

#16

Post by IslandPink »

How long would you like me to talk about this subject, as it's one of my favourite subjects to bore people about, if you have any optics knowledge ? :D
The big companies making the step & repeat machines now are ASM Lithography ( Zeiss lenses ) , Canon and Nikon . [ One of the best designers in the business is David Williamson who works out of the UK for Nikon ].
The current production lenses use 193nm deep UV . They consist of a lens about 1.5m long made up of 30-35 elements made from Calcium Fluoride and Fused Silica , up to 220mm diameter. Essentially it's like a giant microscope objective working in reverse, taking mask designs at 5x final scale and reducing them onto an active chip area of about 30x30mm .
Originally , lenses at 193nm were operating at NA of 0.5 ( = F/1 ) in air, and diffraction-limited. That gave about 90-100nm linewidth, I think. Since then, they have pushed the designs to an NA of over 0.9 in air, then with water immersion between the lens and wafer, to an effective NA of 1.3 - which is how the 40-50nm linewidths have been achieved.

The lens element surfaces have to be finished to an regularity of about 1/100th of a wave of light or better , in the visible . This requires conventional polishing followed by cycles of measurement and ion-beam-figuring ( I believe ) to finish .
The lens elements are mounted into invar ( ~zero expansion) cells and assembled as a stack, one lens at a time, using optical monitoring of the lens to get it centred . The Invar cells are diamond-machined to about 0.5 micron parallelism .

These lenses produce by far the most 'information' in one shot , of any optical systems made in any field . If you wanted to capture the available detail from this lens using a digital sensor, you would need to use about 430 Giga-pixels - I know as I calculated this recently for interest, after seeing an impressive 2G-pix image assembled from shots above Everest base camp .
Alternatively, consider that if the chip was enlarged to 400x400m , the line structures would be at the 0.5mm level .

Very few people understand the extraordinary technology that goes into the common processor chips that are in their computers !
"Once you find out ... the Circumstances ; then you can go out"
User avatar
shane
Social outcast
Posts: 3403
Joined: Sun Sep 16, 2007 12:09 pm
Location: Kept in a cool dry place.

#17

Post by shane »

They're pretty astonishing machines, steppers. Step-and-scan and water immersion machines were still at the theoretical stage when I left. I'd love to have a look at one.

The ASMs were definitely way ahead of the Canons, although in fairness our Canons were much older. To give those who don't know a bit of an idea, ICs are produced by depositing layers of material onto a silicon substrate, coating it with photo-resist, exposing the resist in a thing like a glorified slide projector, then developing it and etching away the original material (or implanting the whole thing in an ion implanter to dope the exposed area of silicon). Anyone who's had a go at making their own PCBs will understand the principal. It's the detail that's astonishing. First of all, as mentioned above, the smallest feature printed on the silicon can be as little as 40nm across. Lets get that in perspective. A human hair (the universal indicator of smallness in the same way that football pitches and double-decker buses are the universal bigness indicators) is about 80 microns in diameter, so one micron is one 80th of a hair. 40nm is four hundredths of one micron, so about 1/2000th of a hair. Features that small have to be printed with perfect definition across a field up to 30mm square. Since the silicon substrate (or wafer, as they're known) we're talking about is up to 300mm in diameter, a grid of exposures is made, with the wafer being moved on a stage under the lens from step to step (hence stepper) until the whole wafer is covered.

That's the easy bit.

Chips are made up from up to thirty layers of material, each one with its own pattern, which of course has to be aligned to the one below to an accuracy of about .01 microns. Think about that. The wafer is 12" in diameter, and is sitting on a stage made of quartz about 15mm thick. That in turn sits on piezo feet that keep the image in focus (depth of field is around 1 micron). This whole assembly weighs about thirty kilos, and has to be aligned under the lens, focussed, exposed, then moved to the next image, aligned again to .01 micron, focussed and exposed in a cycle that takes around one second. To achieve this, the stage sits on an air cushion on top of a lump of granite that weighs around half a ton, and is driven in x and y by a couple of hefty great linear motors, position being measured by laser interferometers. This one-second cycle covers a wafer in about thirty five shots, so a wafer goes through in about 45 seconds, hour after hour, day after day. Astonishing.
User avatar
IslandPink
Amstrad Tower of Power
Posts: 10041
Joined: Tue May 29, 2007 7:01 pm
Location: Denbigh, N.Wales

#18

Post by IslandPink »

That's great info Shane . I know the machines are state of the art in so many ways. Various parts are made out of Zerodur glass-ceramic for temperature stability too I believe.
I was really shocked to hear they had been able to incorporate the water-immersion into all of this to increase the effective f-number - seemed amazing all the wafer-handling and repeat accuracies could work with the additional problems of the water to contend with .
"Once you find out ... the Circumstances ; then you can go out"
Andrew
Eternally single
Posts: 4206
Joined: Thu May 24, 2007 2:18 pm

#19

Post by Andrew »

I think, not 100% sure, they move fast, but 28nm is where its at now.

Andrew
Analogue, the lost world that lies between 0 and 1.
User avatar
shane
Social outcast
Posts: 3403
Joined: Sun Sep 16, 2007 12:09 pm
Location: Kept in a cool dry place.

#20

Post by shane »

Just looking for a video on Youtube to show what a stepper does. Couldn't find anything suitable, but I did find a chap from TSMC talking about their current 28nm process, and plans for 7nm. Isn't that getting down to the dimensions of a single silicon atom?

Couple of interesting bits and pieces:

http://en.wikipedia.org/wiki/22_nanometer

http://en.wikipedia.org/wiki/Immersion_lithography

Is this grabbing anyone? Or boring?
User avatar
Dave the bass
Amstrad Tower of Power
Posts: 12273
Joined: Tue May 22, 2007 4:36 pm
Location: NW Kent, Darn Sarf innit.

#21

Post by Dave the bass »

I think the purple colour is pretty.

DTB
"The fat bourgeois and his doppelganger"
User avatar
ed
retired
Posts: 5384
Joined: Thu Jun 21, 2007 4:01 pm
Location: yorkshire
Contact:

#22

Post by ed »

oh waily waily, the end is nigh I tell you, it will all end in tears.......

you mark my words...oh waily waily.....

no more Moore's law..oh waily waily!
There's nowhere you can be that isn't where you're meant to be
User avatar
IslandPink
Amstrad Tower of Power
Posts: 10041
Joined: Tue May 29, 2007 7:01 pm
Location: Denbigh, N.Wales

#23

Post by IslandPink »

Aha . Looks like they are all still using 193nm exposure wavelength, but pushing the linewidths they can make downwards, by using better and better mask ( original ) design . I'd heard they could shrink the lines below what was really resolvable by the lens, by adding extra 'helper' features alongside the lines on the mask, special patterning . Looks like they can take it a long way .

The next big step was always going to be down to 13nm exposure wavelength in the EUV and multi mirror-systems instead of lenses but I'd heard the light source problems at 13nm were proving to be more and more of a problem and not making progress. There seem to be some comments there that 13nm may never happen . I wonder what's next ? ...

I haven't checked lately but I think a silicon atom is more like 0.1nm , not 7nm , but it does illustrate the nature of the activity a bit .

Dave, I always preferred 405nm myself too .
"Once you find out ... the Circumstances ; then you can go out"
User avatar
pre65
Amstrad Tower of Power
Posts: 21373
Joined: Wed Aug 22, 2007 11:13 pm
Location: North Essex/Suffolk border.

#24

Post by pre65 »

ed wrote:.

no more Moore's law..oh waily waily!
From Wikipedia

"This trend has continued for more than half a century. Sources in 2005 expected it to continue until at least 2015 or 2020.However, the 2010 update to the International Technology Roadmap for Semiconductors has growth slowing at the end of 2013, after which time transistor counts and densities are to double only every three years."
The only thing necessary for the triumph of evil is for good men to do nothing.

Edmund Burke

G-Popz THE easy listening connoisseur. (Philip)
User avatar
jack
Thermionic Monk Status
Posts: 5493
Joined: Wed Dec 29, 2010 8:58 pm
Location: ɐılɐɹʇsnɐ oʇ ƃuıʌoɯ ƃuıɹǝpısuoɔ
Contact:

#25

Post by jack »

shane wrote:Is this grabbing anyone? Or boring?
Actually, I'm finding it really interesting - when I did my EE degree I made some rather rubbish chips - Southampton had pretty good facilities including ion-bean implant machines etc. (in the Zepler building if you know the place) Quite fun, really...
Vivitur ingenio, caetera mortis erunt
User avatar
ed
retired
Posts: 5384
Joined: Thu Jun 21, 2007 4:01 pm
Location: yorkshire
Contact:

#26

Post by ed »

pre65 wrote:
ed wrote:.

no more Moore's law..oh waily waily!
From Wikipedia

"This trend has continued for more than half a century. Sources in 2005 expected it to continue until at least 2015 or 2020.However, the 2010 update to the International Technology Roadmap for Semiconductors has growth slowing at the end of 2013, after which time transistor counts and densities are to double only every three years."
from wiki in Shane's second link

'The ITRS 2006 Front End Process Update indicates that equivalent physical oxide thickness will not scale below 0.5 nm (about twice the diameter of a silicon atom), which is the expected value at the 22 nm node. This is an indication that CMOS scaling in this area has reached a wall at this point, possibly disturbing Moore's law.'
There's nowhere you can be that isn't where you're meant to be
User avatar
shane
Social outcast
Posts: 3403
Joined: Sun Sep 16, 2007 12:09 pm
Location: Kept in a cool dry place.

#27

Post by shane »

When I joined Plessey in 1987 in their brand-new state-of-the-art wafer fab, they were just prototyping their 1.4um process, and thought that the probable limit of conventional photolithography would be around 0.5um. By the time I left in 2004 they'd been running a 0.35 process for three or four years and subcontracted the manufacturing of their 0.18um process to TSMC. By then the supposed theoretical minimum achievable was thought to be around 90nm. Now they're talking about 7nm!

You do wonder where it will all end though. Plessey had three fabs altogether, a 6" in Swindon and a 6" and an 8" in Roborough (Plymouth). Between them they had three Canon steppers and six ASMLs. The Canons were bought second-hand for under a million a piece, but the ASMLs were new at around eight million, I believe. Add to that all the rest of the photo gear (coaters, developers, etc), then add in the etchers, furnaces and implanters, and bear in mind that the whole lot has to be maintained in an atmosphere several hundred times cleaner than an operating theatre, with temperature maintained to within 2 or 3 degrees and humidity equally closely controlled and you can understand that the costs run to the hundreds of millions. Plessey's Fab was interesting because it was small and flexible and was able to develop some really industry-leading innovative processes and get them up and running quickly, but it was always too small to be competitive.

Now look at a company like TSMC who make a massive proportion of the world's semiconductors. They're a foundry, ie they don't design anything themselves, they just make other people's stuff for them. They have one 6" fab, five 8" and four 12", each many times larger than Plessey's total. They have row upon row of ASML step-and-scan machines, which must be around the 30 - 40million mark. According to Wikipedia, they've announced the construction of a new 12" fab for 40nm - 20nm, at a cost of about £10 billion. If they're talking about being able to go down to 7nm soon, then that will be obsolete in three to five years.

If you've ever wondered why you have a drawer-full of redundant mobile phones and yet you still end up getting a new one every 18 months despite there being nothing wrong with the old one, that's why!

What I love about this country is encapsulated by what Plessey are now doing with a Fab that was obsolete 15 years ago and another that was similarly out-of-date in 2004. Just after I left, the 6" fab was moth-balled because there was no real place left for 1um technology when the rest of the world was down to 0.18um. Recently, in conjunction with Cambridge University, they've developed a new process for making high brightness LEDs at about a fifth of the current cost. You don't need nanoscopic technology for LEDs, just some very clever blokes, a lot of blue-sky lateral thinking in your process development and a twentyfive year old 6" wafer fab with a work-force of 80. Oh yes, and to keep the old 8" .35um fab busy, they've developed a process for making electric field sensor chips sufficiently sensitive to be able to detect the presence of a person behind a brick wall.

Bloody clever stuff, and the sort of thing the TSMCs of this world would never come up with in a million years.

http://www.plesseysemiconductors.com/ep ... uctors.php

http://www.plesseysemiconductors.com/ma ... uctors.php
User avatar
Nick
Site Admin
Posts: 15709
Joined: Sun May 06, 2007 10:20 am
Location: West Yorkshire

#28

Post by Nick »

pre65 wrote:
ed wrote:.

no more Moore's law..oh waily waily!
From Wikipedia

"This trend has continued for more than half a century. Sources in 2005 expected it to continue until at least 2015 or 2020.However, the 2010 update to the International Technology Roadmap for Semiconductors has growth slowing at the end of 2013, after which time transistor counts and densities are to double only every three years."
But don't forget Wirths Law:

"software is getting slower more rapidly than hardware becomes faster."
Whenever an honest man discovers that he's mistaken, he will either cease to be mistaken or he will cease to be honest.
Post Reply