E-manufacturing start-ups push computing power to cut mask cycles, improve process control
This year’s Technology Innovation Showcase highlights e-manufacturing start-ups applying big computing power to speed production of good photomasks, and using clever software integration to test assembled gas delivery systems, and to combine multiple suppliers’ sensors into one-fault detection system at SEMICON West. All of these jury-selected winners will be highlighted in the Manufacturing Productivity and Effectiveness TechXPOT, Esplanade Hall.
In the design for manufacturing (DFM) field, Luminescent Technologies and Petersen Advanced Lithography are pushing computational lithography to radical new extremes to significantly shorten the time it takes to make advanced photomasks that print.
Luminescent Technologies, Mountain View, California starts with the desired circuit pattern, adds the characteristics of the scanner and resist, then calculates back to devise the best mask pattern to print the desired image with the largest process window. “It comes up with the mask that will print the best, without the army of programmers to move the edges around,” says Andrew Moore, vice president of corporate marketing, noting that increasingly complex traditional optical proximity correction (OPC) work is pushing chip makers to start to ramp from 10 up to 20, or from 30 up to 50 programmers. Key to the process’ efficiency is a dedicated massively parallel hardware system that speeds up running the image-based program—and its ability to optimally place the scattering bars at the same time as generating the basic mask pattern.
The Luminizer can create highly detailed mask patterns with nearly continuous contours that get good yields and make impressive presentation slides; but that, critics say, would be tough to manufacture. Probably more practical will be using it on less detailed settings for most of the pattern, with larger fracture sizes of 30 nm–40 nm that look and act a lot like regular OPC, and using a slower, higher detailed setting that gets the better process window only on the critical problem spots. Moore says the company showed at Photomask Japan that they could usually find enough suitable structures in the design’s L and T patterns to stand in for parallel bars while inspecting the masks. Repair remains an issue, since the mask feature sizes needed to get the best yields are smaller than the size of the ion beam of the repair tool, though new AFM repair tools can do better. “But our competitors will face the same issue,” argues Moore. “Mask reparability is an industry-wide issue.”
Moore says the company has shown that its inverse computation can get the same or better results as some chip makers’ internal OPC methods, but in much less time, noting a case of one non-lithographer creating all the critical layers for a 45 nm flash chip in only two man-weeks. The company is in its second silicon turn with some chip makers evaluating the full-chip version of the technology. Cypress, SMIC, UMC and Xilinx have all presented papers with the company in the past.
Latest iteration of John Petersen’s lithography solution is a Maxwell equation solver that calculates the impact of the electromagnetic field on the mask pattern over large areas (4000 nm and larger) to adjust the OPC to get the pattern to print as desired, and pay-per-use access to the massive computing resources needed to run such analysis. Petersen Advanced Lithography, Austin, Texas, has now started marketing the system in earnest, as a kernel for OPC companies to plug into their own products, as standalone software, or in a turnkey system; and is working with KLA-Tencor to integrate it into its PROLITH simulation tool via PAL’s Image Design Factory environment. But fastest growth currently is coming from its utility computing sales, for doing computational lithography on the company’s massive parallel cluster for PROLITH and an associated supercomputer for EMF3. “These are huge computing problems, requiring equivalent computing power to that needed to solve the human genome,” says President John Petersen. “But the era when this kind of computational power is needed is just beginning. Later when it becomes more common, companies will probably invest in their own hardware.”
Petersen argues that inadequate fundamental understanding of the interactions between the electromagnetic fields, the light and the mask materials forces companies to rely on iterations of trial and error to figure out how the mask pattern will get distorted. “OPC now uses assumptions about the electric field,” says Petersen, “But the OPC is wrong.” He notes that contact holes may still show up to 90 percent variation from the target size after OPC; but, after adjustment with the Maxwell solver will show almost no variation. “Anyone who is using non-printing assist features in OPC is having these problems. At 45 nm they will be pervasive. And the problem explodes with immersion, as the more extreme angles lead to more extreme polarization,” he asserts. “We fully believe we can eliminate at lease one cycle of value, and likely two.”
In the advanced process control sector, the Technology Innovation Showcase Committee highlights software that enables the testing of assembled gas delivery systems, and the integration of data from multiple sensors from different suppliers into a single fault detection system.
Seaware Technologies, Milpitas, California aims at the underserved niche of test and diagnostic work for gas and liquid delivery systems. “There is no way of testing the entire system when all the components are assembled,” says CEO Scott Hepworth. “Testing is mostly manual and labor intensive … There is no independent way of getting actual flow and pressure measurements to identify variances.” So his start-up aims to integrate hardware and software into a tester that can, and to sell that testing as a service for a monthly fee. First target is final bench testing for gas box manufacturers, and the company has just finished beta testing at one supplier. Hepworth says one study showed the system cut testing time from 35–50 hours down to 7–10 hours, with only ½–1-½ hours of that actually involving the technician’s time. “We have lots of support from manufacturers of mass flow controllers,” he notes. “Because they’ve typically been the scapegoat when people can’t find the leaky valve that’s really the problem.”
Straatum, meanwhile, now makes it possible to incorporate multiple different sensors from different vendors into its fault detection and control system; integrating them into its software and algorithms, to make multi-sensor fault libraries in the central fab database. Main target is to add optical emission spectroscopy for plasma etch for better sensitivity to the chemistry to the company’s own radio frequency sensors that are most sensitive to hardware and chamber issues. “RF has limited sensitivity to chemical changes, but the optical emission spectroscopy sensors increase sensitivity to minute gas changes and drifts, and give a good indication of what results will show up in metrology 12 hours later,” says CTO Marcus Carbery. The system can also integrate downstream in situ particle monitors and other sensors. “Generally these are from small start-ups with great sensor technology, but without truly scalable software and without the resources to develop it,” says Carbery. The product being introduced commercially at SEMICON West has been beta tested at two big customers, one in the U.S. and one in Korea.
For more information or to register visit www.semi.org/semiconwest.