IBM unveils worlds first 5nm chip

How many silicon atoms does it take to make 5 nm?

George H.

About 45.

If you want a 5 nm long "wire" you need about 9 lattices, each with 8 atoms.

ms.

Right, (not a well formed question) Say a square wire, 5nm on a side. Then (as you say) 9 lattices times 8 atoms / lattice, a total of

George H.

On a sunny day (Mon, 5 Jun 2017 11:52:42 -0400) it happened rickman wrote in :

Yes, google gives 210 pm (pico meter) for a si atom, so

-> 5E-9/210E-12 = 23.8095 so 24 without the big atom splitter :-)

But I am no chemist .. likely there needs to be some crystal structure or something. Maybe Bill Sloman?

One scanner, with its EUV source, costs over $120 million. The NRE for one chip, design and masks, is about $250 million. Not many chips can justify that: flash, DRAM, possibly a high-volume processor.

John Larkin wrote on 6/5/2017 1:27 PM:

Did you see the size of the chip in the link? It's larger than any chip I've ever seen. I think 2x4cm.

Silicon forms a face-centered diamond-cubic with a 0.543 nm side length and 8 atoms per cube, see:

so a 5x5x5 nm silicon cube contains about 9*9*9*8 = 5832 atoms

Den mandag den 5. juni 2017 kl. 19.28.08 UTC+2 skrev John Larkin:

and fpgas

Maybe. The trend seems to be to fab one chip and sell it as a range of parts, by blowing fuses or by selling partly defective chips as cheaper parts. We buy two different-size, different-priced Altera parts that seem to be identical.

Here's an EUV scanner. The EUV target droplet chamber is at the far end.

I don't know why those guys are wearing bunny suits. The wafers are in vacuum.

ISO 9000

On a sunny day (Mon, 5 Jun 2017 20:08:33 +0200) it happened Miguel Giménez wrote in :

Thank you, nice visualisation.

John Larkin wrote on 6/5/2017 2:51 PM:

The same reason you don't pile the hospital waste in the room next to the OR.

John Larkin wrote on 6/5/2017 2:51 PM:

It's not that they sell "defective" chips as smaller chips. They save money by simply testing less of the chip and selling it as a smaller device. Just like for very high volume customers who aren't looking for reconfigurability, they will test to your design usage which ends up being a

Or high value processor. Are you sure about the $250M number?

Could equip the next several of these

NT

Precisely. As my boss at IBM once said, it's better to work for people with money. ;-)

That's a number that I've seen in electronics mags.

It looks like those 2x4cm structures might be several dies with a test die for the group.