Intel CEO Pat Gelsinger is confident that his company will outmaneuver TSMC in the coming years. In an interview with Barrons’, he highlighted the strengths of the 18A process (1.8nm) versus TSMC’s N2 (2nm) node. Both 18A and N2 will leverage GAA transistors (RibbonFET), the successors of the widely-used FinFET technology. Intel’s 1.8nm-class node will also feature PowerVia, a backside power delivery technology that optimizes power and clocks.
Intel has already tested PowerVia on its 4nm-class node introduced with Meteor Lake. Per the chipmaker’s internal tests, PowerVia demonstrates >5% frequency improvement and >90% cell density on Intel 4. The 18A node won’t be the first to leverage RibbonFET or PowerVia. That honor will go to the 20A (2nm) process launching with the 15th Gen Arrow Lake processors next year.
We announced two major innovations with 18A: a new transistor and backside power. I think everybody’s looking at the transistor of TSMC’s N2 versus our 18A. It’s not clear that one is dramatically better than the other. We’ll see who’s best.
But the backside power delivery, everybody says Intel, score. You are years ahead of the competition. That’s powerful. That’s meaningful. It gives better area efficiency for silicon, which means lower cost. It gives better power delivery, which means higher performance. So, I have a good transistor. I have great power delivery. I think I’m a little bit ahead of N2, TSMC’s next process technology in time.Via Barrons.
Intel may have the lead on a technological level, but what it doesn’t have is ample production capacity or a good track record. Despite multi-year delays, its 10nm node was limited to the mobility market for two generations, indicating limited production capacity. The same can be said for the Intel 4 (4nm) node powering Meteor Lake.
The Core Ultra processors show a sharp regression in clock frequency with minimal efficiency gains. The availability of these chips outside the US is also low. TSMC has a long list of devoted customers whose roadmaps rely on its highly proven execution strategy. These include AMD, NVIDIA, Qualcomm, MediaTek, etc. On the other hand, Intel can’t produce 4nm-class Meteor Lake chips that are as efficient as AMD’s year-old Phoenix parts.