Is better OPC an alternative to immersion lithography?

With the notable exception of Intel, most fabs have announced that they will be using immersion lithography for critical layers, certainly by 45 nm. This represents both a major capital investment and yet another yield learning curve, as immersion techniques require a good deal of work to master. Now it appears that for many pattern sets, immersion is not the only alternative.

Luminescent Technologies has recently announced progress and some production results for their novel inverse lithography transform technology. (Please see Ann Mutchler’s excellent story here.) The company claims that initial customers have been able to employ conventional dry lithography on layers that previously required immersion, simply by using Luminescent’s tool to generate the mask data.

The tool works by creating, in effect, an inverse transform of what happens to the mask image between the time it is written on the mask and the time the etch process has turned it into patterns on the photoresist layer. By feeding the patterns that you want to have in the photoresist into the inverse transform, in theory, you get out the patterns you need to put on the mask. The inverse transform takes into account optics, dose and focus variations and the vagaries of the development and etch steps.

In practice it’s not quite that simple. There isn’t a one-to-one mapping between mask and resist images: there are often many different images you could put on the mask that would give you approximately the right shapes on the wafer. The choice varies with your sensitivity to dose, focus and other variations, and to the practical limitations of your mask geometry. Hence the inverse transform process must be guided by heuristics.

The Luminescent results open up all sorts of interesting questions. For example, how much of the investment in next-generation lithography could be delayed by one or two process nodes if such techniques are used? Is the technique applicable beyond the lithography and etch steps, to reduce variations in diffusion, well-proximity effects, line-width variations and so forth—that is, to reduce process variations? If Luminescent has achieved significant results already, can more work yield dramatically better results? Are captive fabs already using similar techniques in their internally-developed tools?

Lots of questions. What seems to be beyond question is that the Luminescent results have opened up a new axis in the already multidimensional problem of process control and optimization.