EUV progress

On a sunny day (Wed, 27 Sep 2017 17:45:46 -0700) it happened John Larkin wrote in :

Sometimes I think the quest for ever smaller is similar to teh one for to ever less distortion in audio amps in the sixties. Sure more chip memory... needed by bloated OSes with bloated software written in bloated languages by people who have no clue of the hardware.

US went to the moon and back with less computing power and memory.

It is a make believe world. on top of that any co[s]mic ray will utterly destroy the stuff. And on top of that there is an atom size limit and noise.

It is about competition and market share and not about what is really needed.

The light sources are ... pricey, uncertain. Next generation, just might use synchrotron sources instead (can't make 'em small, but it's better proven technology).

There's outliers, of course; IBM has demonstrated atomic-level assembly, and a track-following nanobot array might actually build soething useful... someday.

Fabrication speed is the issue. Electron beam microfabricators can write very fine lines, but too slowly to make much money.

Bell Labs did introduce and sell a "shaped beam" electron beam microfabricator, and European Silicon Structures bought ten of them

Sadly, what Bell Labs claimed about the writing speed their machine could manage, and what ES2 could get out of them differed by a factor of ten, and the company went bust.

I wondered what happened to them; by 1988 I'd lost what little interest I had in that sector.

I never did understand how they expected the speed to scale with (very rapidly) increasing transistor count.

e:

ight

y,

e very

uld

and

had

The transistors got more numerous because they got smaller. E-beam is area limited, rather than feature limited (up to a point) so writing lots of sma ll transistors in the same area as a few larger transistors isn't going to take much longer.

The Cambridge Instruments shaped beam machine I worked on - which never mad e it to the market, mainly because Thompson-CSF sold us a proof-of-principl e machine which they thought was a pre-production prototype, and we initial ly believed them - could write rectangular shapes, which could be up to 10 microns long or wide - but not at the same time. 10 microns by one micron(o r 3 microns by 3 microns) was as much area as we could illuminate in a sing le flash.

...at least ONE person is awake.

Fabs building the finest-pitch semis are extraordinarily expensive, so they're only practical for making high value, high volume chips. Storage-class flash memory, for instance.

The individual transistors stopped improving with the 65-nm node, and have been getting slower, leakier, and flakier ever since, despite a lot of work by a lot of smart people backed with a metric buttload of money.

Cheers

Phil Hobbs

What happened to the cosmic ray thing anyway? I remember people worrying about it back in the 80's when the transistors were like, what, a million times "bigger"!

kin

Perhaps error-detecting and correcting codes. When we wanted to store 64-bi t words back in the 1980's we planned to buy 72-bit wide storage, and add e ight check bits per word, which would have caught and corrected every singl e bit error and most 2-bit errors.

On a sunny day (Fri, 29 Sep 2017 15:15:59 +0100) it happened John Devereux wrote in :

It is a serious problem, and not long ago there was (here?) discussed a paper about the effect of co[s]mic rays on airplane electronics at high altitude. bad bad bad. I had a discussion online some years ago (2?) with somebody who wanted to keep anonymous but turned out to be from Intel ;-) about the problems with FLASH, they had BIG problems.

I have a nice co[s]mic ray detector setup here, you would not believe the stuff that hits it at sealevel ..

PS: This is from year 2000 or so, even then 5 fold backup electronics was used in the space shuttle. with much lower density SRAM: wget

Just use Bing to search..

Small feature-size processes are LOUSY for high quality analog functions. I just finished a design on 180nm. Leaky CRAP! ...Jim Thompson

Hard drives kept pushing density up until the error rate went up, then added error correction and RAID tricks to recover from the errors. Flash storage does that sort of thing too. I suppose that if you had a flash or DRAM technology with a 40% bit error rate, you could still make it reliable with ECCs and redundancy.

EUV is so expensive (120 megabucks per scanner, a quarter billion for design and mask sets for one chip) that not many chips will use it. But those chips will sell in the billions.

Walking from work to the parking garage yesterday, 80% of the people walking on the street were online somehow. Drivers stopped at red lights mostly had their heads down. The cell bandwidth and storage requirements are awesome.

You probably don't need twenty billion transistors on a chip either.

What's your detector?

This is fun. (muon lifetime)

I think the typical muon/particle flux is a few per second. (for a scintillator that is about the size of a loaf of bread.)

George H.

Which makes the acquisition of Toshiba's memory business an interesting bet.

Bain is Mitt Romney's thing, isn't it?

Perhaps but there was a need for "radiation hardened" stuff in aerospace even in the 80's but nevertheless I think you are happy enough with your pis?

There's certainly a lot of that in flash and dram AIUI. Don't know about the big CPUs though.

Nope. ...Jim Thompson