Here's how the company wants to one-up its main rival

Oct 31, 2014 07:36 GMT  ·  By
An image of a die that was given a bevel polish, rendering the transistors visible
5 photos
   An image of a die that was given a bevel polish, rendering the transistors visible

Intel's main competitor on the central processing unit market may be AMD (the latter's wish to free itself from the comparison notwithstanding), but when it comes to the actual foundry business, the CPU tech manufacturing and development, it's TSMC that Intel has to contend with.

Intel makes its processors itself, which means that it possesses its own fabrication plants, where its plans for newer and better chips are put into practice.

AMD had its own foundry for a while, but then sold it off, and Globalfoundries is still young and untried. Also, AMD, like NVIDIA, still orders its GPUs from TSMC.

Taiwan Semiconductor Manufacturing Company is the biggest worldwide foundry (another word for semiconductor production plant).

Intel and TSMC have different roadmaps for transistor size advancement. Different view of Moore's Law. And their chosen paths mean that sometimes the manufacturing node (in terms of nanometers) isn't enough to judge whose chip is better. Especially since said chips are usually made for different things.

Because of that, it wouldn't do to assume Intel's 14nm Broadwell CPUs would be clearly technologically superior to TSMC's 20nm technology.

Intel claims that to be the case regardless

Intel announced details about the 14nm process as early as last year, saying that it would be a huge leap in performance, efficiency, endurance and everything else. It even clearly said that the 14nm Broadwell would be better, smaller and less power hungry than any TSMC 20nm chip.

Now, the corporation's claims are finally being verified by independent engineers from Chipworks. Their conclusion is that Intel really is putting its money where its mouth is.

The Broadwell has much better transistor fin pitch (around 42nm), transistor gate pitch (70nm) and interconnect pitch, plus a reduction in SRAM scaling. The processor also has 13 metal layers instead of the previous nine (11 for Bay Trail). The layers connect different areas of the chip, and doing this efficiently affects the processor performance, since it influences the behavior of the transistors.

The more metal layers are used, the harder it is to make the chip efficient, but Intel managed it somehow for Broadwell, although the interconnect pitch of 54nm is a bit larger than the promised 52nm (though still within the margin of measurement error according to the company). The SRAM cell target size is of 0.058 µm2 for those interested.

What all this means in layman's terms

The Broadwell chips, which got introduced with the Core m ultra-mobile processor and will be found in the 2015-bound desktop/laptop CPUs, are definitely much stronger compared to Haswell, all the while using as little energy as what passes for a laptop and even tablet processor right now. As long as the chips are properly coded anyhow.

Update October 31, 2014: Fixed the mention of AMD and Globalfoundries, as the former sold off the latter a while ago.

Chipworks reverse-engineering shots of 14nm Intel Broadwell (5 Images)

An image of a die that was given a bevel polish, rendering the transistors visible
Cross-section showing the thirteen metal layersThe edge seal renders 12 of the 13 layers easier to see
+2more