News Nvidia and Broadcom continue trialing Intel 18A test chips: Report

[Admin]

A new report suggests that Broadcom and Intel are continuing to test the Intel 18A manufacturing process, amid reports of setbacks for contract manufacturing customers.

Nvidia and Broadcom continue trialing Intel 18A test chips: Report : Read more

 

[MXM0]

I believe the TSMC N2 SRAM density claims were recently disproven in a trade show, where 18A and N2 were shown to have nearly the same density (iirc there was a ~3% difference, but I don't remember which was "better").

18A does, however, have the advantage of backside power, which, presumably, reduces it's power requirements by about 20%, over N2, and hence doesn't have as many thermal issues, if the chipmaker wants to save power, or allows higher input if they want to be faster, etc.

 

[Mr Majestyk]

TSMC will have a more advanced type of BSPD in the A16 node than Intel is using. Intel's approach is the middle level complexity of 3 approaches to BSPD and will be a big breakthrough though in itself. But will we even see Panther Lake this year? Clearwater Forest delayed a year until 20206 sometime, Panther Lake pushed back to late Q4 and IMO will now be a paper launch in 2025 to keep investors happy.

 

[MXM0]

TSMC will have a more advanced type of BSPD in the A16 node than Intel is using. Intel's approach is the middle level complexity of 3 approaches to BSPD and will be a big breakthrough though in itself. But will we even see Panther Lake this year? Clearwater Forest delayed a year until 20206 sometime, Panther Lake pushed back to late Q4 and IMO will now be a paper launch in 2025 to keep investors happy.

A16 isn't due until like 2027. They are a long, long way behind Intel, with backside power.

Moreover, TSMC outright stated that they will not buy a EUV machine until 2030, and will, instead, continue to use DUV multipatterning for N2 and A16. That will yield horrible failure rates , as multipatterning always does.

Why do you think nVidia GPUs are so expensive, these days? Or why they ship GPUs with disabled ROPs? You think they don't test them? It's likely that the failure rates of TSMC's modern processes is just that bad.