For Lithography, Random Events Could Dominate and Doom

A common topic at the SPIE Advanced Lithography symposium concerns random events, or stochastics. Controlling these or lessening their impact on device yield or performance is critical to continued semiconductor advances. But there are signs that the small feature sizes and large number of transistors on future chips may make this too difficult and perhaps impractical to do.

An example of the seriousness of stochastics was found in a Tuesday, 26 February presentation on EUV given by Eric Hendrickx, an IMEC program manager in patterning. extreme ultraviolet (EUV) at 13.5 nm wavelength is the next step in lithography.

Researchers at IMEC investigated EUV lithography printing problems. Summarizing the findings, Hendricks likened it to sailing a ship. "If you're between two cliffs, you may not have much room to navigate," he said.

The study turned up two opposing cliffs that threaten manufacturing viability. It also found a floor that could make manufacturing too expensive.

In the study, the IMEC team did tests using an array of lines or holes, also called contacts. The first can bridge, causing what should be separate lines to touch. Lines also can break, another failure mode. As for contacts, they can remain closed, failing to open, or they can bridge. Again, there are two distinct failure modes.

Some of these problems arise from random events. The number of photons in an exposure shot fluctuates stochastically. Another source of randomness is in the resist, which is full of molecules that react slightly differently to those photons. Etch processes that remove film layers are also somewhat random.

In the test, the researchers varied the array spacing, starting with a 40-nm pitch for contacts of varying size, for example. They then counted how many cases there were of closed or bridged contacts. Because tomorrow's chips may have hundreds of billions or even a trillion contacts, the scientists extrapolated from their much smaller sample as to what would be the case for an actual chip.

What they found is that the chance for closed contacts decreases as size goes up. However, increasing size makes the probability of bridged contacts go up. So, the process sits between two opposing requirements - or essentially between two cliffs. What is needed is a window where both defects are in the parts-per-trillion range. According to the IMEC scientists, that may not be possible unless the array spacing is above a process-dependent threshold.

They also observed a floor, a minimum density of printing problems that did not seem to respond to changes in contact size. This, they theorized, might be due to random problems in the resist, although they could not rule out other possible sources.

The array of lines presented a similar situation. This provides chip makers with a means to determine if a design is manufacturable using EUV. They can start by looking at the most congested part of a chip and then measuring the print failure rate, according to Hendrickx.

For the industry, the goal is to separate the two cliffs and reduce the floor. That will create space for manufacturing.

However, in a presentation on Monday, 25 February, Chris Mack, chief technical officer at Fractilla, indicated that stochastics could pose a significant problem. His contention is that stochastics-limited lithography forces some unpleasant tradeoffs. This is because doubling resolution, such as would be the case when moving from the 14 to 7 nm feature size technology node, requires an eight-fold increase in write time. If true, that would quickly bring an end to feature size shrinks because the manufacturing time and cost of the new process nodes would rise too quickly.

Because this behavior is rooted in physics, it could apply to all nanolithography techniques dominated by random events. If so, overcoming these effects could be difficult, said Mack.

"The knobs available for us to turn to make this better are not that good," he said.

The demise of lithography-driven feature size shrinkage has been predicted before. So far, ingenuity and large investments in research and development have always found solutions. Only time will reveal if EUV is as successful and avoids the stochastic trap.

Hank Hogan is a science writer based in Reno, Nevada.

See more news and highlights from SPIE Advanced Lithography.