by Katherine Derbyshire, Contributing Editor, Solid State Technology
One of the most powerful process optimization tools lithographers have is the focus-exposure matrix. It cycles through a wide range of focus and exposure parameters, defining the process window quickly and efficiently, with minimal loss of valuable production stepper time.
In the pharmaceutical industry, researchers use automated simulation and synthesis tools to create and test thousands of related compounds. Only a few are likely to be medically interesting, but this combinatorial approach identifies those few in a fraction of the time required by other methods.
In most semiconductor fabs, however, development engineers must fight for every minute of equipment time. Manufacturing engineers reject any experiment that might contaminate the process of record. Testing more than a handful of “safe” candidate materials is difficult, exploring the full process space is nearly impossible. At the same time, the number of important process variables is exploding. Fabs are using more different materials than ever before, dramatically increasing the number of important interactions and the process parameters needed to control them.
According to president and CEO David Lazovsky, Intermolecular hopes to help fabs use combinatorial methods to explore larger process spaces more thoroughly, with its Tempus R&D platform. One of the biggest obstacles to that goal is the silicon wafer itself — 300mm wafers, and the equipment used to process them, are simply too big and too expensive to waste by screening low-percentage process and material candidates.
Instead, Intermolecular equipment isolates hundreds of sites per wafer, applying a different cleaning, electroless deposition, or other wet chemical process to each. Thousands of candidates — which can be new materials, new process conditions, or combinations of both — can be screened with a single lot of wafers. Secondary and tertiary screening using larger isolation spots can narrow the possibilities further, extracting the most promising candidates for conventional full-wafer testing. Supporting software generates the test chips and test recipes used to screen candidates, depending on the specific process question being investigated.
Such large-scale experiments produce an enormous amount of information. Even process conditions that are not appropriate for manufacturing help characterize the system, and may prove useful if other parts of the integration scheme change. Intermolecular’s second major development is a system of informatics software that can log, analyze, and archive thousands of process candidates.
Since such large scale experiments have historically been so difficult, it’s hard to predict what fabs could do with so much information. Early results hint at intriguing possibilities. One project performed more than 7600 experiments on 60 different candidate molecular masking layers for copper interconnects, Lazovsky said, generating more than 18,000 different sets of characterization data. The project identified two process/material candidates for electrical testing. That seems like an enormous investment for very little return, except that the project consumed only one lot of wafers and took four engineers just five weeks. Development of a high-volume manufacturing process based on one of the materials took just eight months from the beginning of the project, and consumed less than 50 wafers.
It sounds too good to be true. Yet the same methods have established themselves in a wide range of other industries. If Intermolecular’s tools provide the degree of isolation the company claims, they can help shatter one of the most significant barriers to cost-effective process development. — K.D.