Litho Is Out Of Sync

EUV’s repeated missed deadlines, and the slow-motion response by the rest of the industry to fill the void with alternatives, is having ripple effects in every facet and corner of the semiconductor industry. It’s making design harder and more expensive, introducing potential errors into the DFM flow, and greatly increasing the amount of time it takes to process wafers. It’s also adding a level of uncertainty that is keeping companies at 40nm and 28nm, and pondering which route to take next.

All of this adds to the already skyrocketing costs and limits innovation. Moreover, it doesn’t appear as if things are going to get better anytime soon. An engineer who left the industry six years ago and returned commented that things haven’t changed much—it’s still all about EUV, shrinking features, and slicing down costs. He has a point.

Put in perspective, planning on the manufacturing side historically has been stellar. In fact, it could well be the most efficient planning in the history of business, allowing us to buy computers that are more powerful than the early PCs that we can wear on the bridge of our nose, inside our bodies, or carry in our pockets. The number of pieces that have come together at just the right time, and over time, probably qualifies as a benchmark that can be used for MBA students learning about ecosystems and how to cut costs consistently over time.

This isn’t just some random glitch in the Matrix, though. What’s surprised everyone outside of a couple of companies is just how little we know about light and just how unyielding its properties are—particularly how to speed it up using some mind-boggling power source technology. And no matter how many brilliant scientists and billions of dollars we throw at the problem, it’s still only partly solved. EUV will continue to play an important role for many years and at many process nodes, but what’s becoming clear is this is a problem that won’t be solved with brute technical know-how and engineering bravado. It will require additional work on EUV alternatives—multi beam and DSA are the most promising at this point—in addition to a rethinking of manufacturing flows because not all parts need to use the most advanced lithography.

Had EUV met its deadline this rethinking of the entire litho problem still would have been required at some point. But it would have eliminated the need for double, triple, quadruple and now octuple patterning at future nodes. Still, by falling out of sync, it has forced other parts of the design through manufacturing ecosystem to start thinking well beyond the normal process of shrinking everything at once using the same lithographic approach. There is much more interest in the past six months in stacked die technology, and there is more effort and resources being invested in 28nm to make that node last much longer than previous nodes.

The bottom line is that lithography has shaken up the entire industry, and good things usually come out of that kind of disruption.