Last week’s news that IBM Corp, AT&T Corp, Motorola Inc and Loral Corp are discussing teaming up on X-ray lithography and regard it as the way forward for chip fabrication (CI No 2,467), came as little surprise to Peter Williams, chairman and chief operating officer at Oxford Instruments Plc. Oxford supplied IBM Corp’s Advanced Semiconductor Technology facility in East Fishkill, New York with a Helios synchrotron, the doughnut-shaped electron accelerator several yards in diameter, in 1991, and Williams says it was common knowledge that the four had been living in sin for years. For Williams, the belated leaking of the news ups the stakes a little bit and places his company in the very pleasant position of being the only viable game in town. The victory of X-ray lithography has always appeared certain but there have been many casualties along the way and optical lithography has failed to roll over and die at the 0.5 micron barrier as the script demanded. Since 1986 Computergram has been running stories about the technique that uses X-rays rather than ultra-violet rays to etch circuits on silicon chips. The advantages from a performance perspective of X-ray over ultra-violet appeared overwhelming.
Wildest dream
The precision of the etcher is inversely related to its wavelength, and as X-rays have a wavelength about 1,000 times smaller than light, they can draw much finer features. Also we were told that ultra-violet rays could not be focused below 0.5 micron, because diffraction started to get in the way, whereas X-rays only started at the 0.5 micron level and could be honed down to 0.001 microns, far beyond the wildest dreams of microelectronics. Thinner wires mean you can cram more components onto the wafer and this not only speeds up the chip as components are closer together, it also means that you can increase the production rate of chips as you fit more chips onto the 8 wafer, without having to up the throughput of wafers in the fab. However, the optical lithography dinosaur fought back. Heinz Hagmeister, chairman of Europe’s Jessi research programme admitted to Electronics Weekly in an interview in February that: We thought that 0.5-micron would be the limit for light but we were absolutely wrong. The optical people did a much better job than the scientific community expected, and now everyone agrees that up to and including 0.18 micron lithography – the level required for a 1G-bit memory chip – will be optical, whatever the source of light. This would probably be the full extent of optical lithography, as its difficult to make lenses for the shorter wavelengths of ultraviolet, because practically nothing is transparent to them. But this step should keep the technology afloat into the next century or three more generations of chips, and as yet very few chips have been successfully cut with X-rays – an IBM memory chip and some simple logic circuitry from Nippon Telegraph & Telephone Corp. Proponents of optical lithography have also pointed to the teething problems of the X-ray technique. In lithography you require a stepper that projects the required pattern onto the Silicon wafer. The stepper used in optical lithography is made of blank quartz and then Chromium is used to trace the design. It uses lenses to focus the beam through the chip mask onto the wafer. The mask can be therefore five times larger than the chip and thus easier to cut. By comparison, focusing X-rays is incredibly difficult and so the X-ray stepper mask has to be the same size as the chip to be cut, and this has been cited as a problem. The stepper relies on casting a sharp X-ray shadow of the mask on to the wafer, rather than a focused image. According to Williams, cutting is not a problem as electron beams can be used to cut the mask accurately, and whereas any dirt on a optical mask will stop the light shining through and ruin the chip, X-rays travel through dirt and therefore yields from X-rays will be higher. Detractors of X-ray lithography have suggested that another problem with the miniaturisation is that a chip can have up
to 32 fabrication layers, and all the masks must be exactly aligned, which is harder with smaller masks. However masks can now be aligned to within 100 Angstroms on the X, Y, and Z axes. Other methods have been touted as alternative successors to optical lithography.
By David Johnson
Electrons can be focused down to the same wavelength as X-rays, and can be used to write features directly on the chip under numerical control, but the technique has failed to excite the market because a single electron beam takes far too long to trace out the millions of features on a chip. However now people are talking about using several beams simultaneously on different parts of a chip at the same time, thereby accelerating the process. Williams regards this as a non-starter even well into the next century and the only place for electron writing is for the masks for either optical or X-ray masks. With optical and X-ray lithography you flood the surface of the silicon with rays for between 0.1 and 1 second. This is far quicker than the time it will take to write a 1,024 by 1,024 pixel grid with one, 10, or even 100 single electron beams and the engineering feat required to control them is mind-blowing. Similarly the idea of bundling together tapered optical fibre to channel X-rays to a point, a technique invented by Russian physicist Muradin Kumakhov and enhanced by the American Walter Gibson, is a series of technical nightmares: sharpening the fibres to a 0.2 micron point and combining and controlling the number of fibres necessary to attain speeds comparable to X-ray. Ion beam lithography is even more far-fetched, and according to Williams is into the realm of white haired scientist, as you fire the beams into gaps in space rather than gaps in a mask. Moreover there have been problems in developing the synchrotron and Oxford is one of only two survivors of a catalogue of failed ventures. In 1991 Grumman Corp’s Electronic Systems Division’s attempt foundered, having swallowed huge quantities of federal dollars with no return. It managed to build a conventional magnet mock-up but its partner, General Dynamics Corp, failed to supply the necessary superconducting magnets. Oxford had the necessary magnets and attempted to bid for the contract, but was barred by the US regulators. A similar fate swallowed the COSY-MicroTec project of Leybold-Heraus AG of Cologne back in 1986. Here state regulations allowed Oxford to design the ring but not to bid for construction. A subsidiary of Siemens AG struggled manfully but vainly with the problem, and the project collapsed. In 1992, Oxford was allowed to bid for the supply of a synchrotron for Louisiana State University. Unfortunately, the senator for Louisiana, Bennett Johnson, also happened to be the chairman of the Senate Energy & Commerce Committee, and unsurprisingly Oxford failed in its bid. The contract went instead to Maxwell-Brobeck Inc of Richmond, California, and Williams concedes that Louisiana has a very good facility. However Maxwell-Brobeck went in at a very low price and went bust. In Japan, the Ministry of International Trade & Industry, in conjunction with 13 private companies, set up the KK Sortec consortium in 1986. Sortec was given $140m in funding, 70% from MITI, but unfortunately used it all up in building a full scale national laboratory accelerator and hence had no money to develop lithography techniques. It has gone back to state for more, but the economic position of Japan means more funding is not a foregone conclusion. The protectionist approach that governments have taken over X-ray lithography has been a mixed blessing for Oxford Instruments. If sensible market considerations had triumphed over nationalism, Oxford would probably have sold more than one synchrotron. On the other hand, governments’ policy has effectively killed off any competition.
120 tons
Hitachi Ltd built the SuperAlis synchrotron for Nippon Telegraph but has no intention of marketing it because it would be too horrendously expensive. Oxford’s only rival, the Aurora from Sumitomo Electric Co Ltd
weighs in at a less than manageable 120 tons, it’s iron clad, compared with Helios’s trim 28 tons, and offers much lower throughput because of lower maximum beam intensities. The significance of last week’s news is that four major players are assigning the future of chip manufacture to X-ray lithography, and quashing the optical’s fightback. The move will also benefit Oxford, and makes the selling of its second Helios much more likely, particularly to a Japanese company. Williams says there are several potential customers, but between potential and realisation there exists a gap. Intel Corp is the big stand-out, having committed itself to optical lithography for its next generation of chips. Williams warns against reading too much into this announcement as Intel has a history of reaping the rewards of other people’s investment and will rely on buying tools in the open market. On the other hand, Mitsubishi Electric Corp recently said it would use X-ray lithography to cut its 256M-bit chip, and is building its own very large and very expensive synchrotron. For Williams, the size of the market for the synchrotron is not dependent on the acceptance of the technology, that’s a foregone conclusion, but the number of chipmakers to survive into the next century – and this Williams believes will be in single figures for making the 1G-bit chip – but the companies consortia – that do survice will have more than one plant, and each of these will require its own source of X-rays.
