18 Jun 2004
Illuminating silicon wafers with a pattern of light has long been the process of choice for making microelectronics, but there are fears that it will not be able to meet future demands. Steven Brueck, a supporter of the technology, argues why it is here to stay.
From Opto & Laser Europe June 2004
Moore's law, the two-year doubling of the number of transistors in an integrated circuit (IC), is ingrained in our expectations of technological progress. In fact, the communications and data-storage sectors have their own "Moore's laws", describing exponential increases in capability. These laws are measured in terms of the bandwidth of optical fibre and the density of data storage in bits per centimetre squared, respectively.
Underpinning this immense progress in the world of information technology is a continual improvement in lithography which enables the creation of ever smaller features inside semiconductor chips and optical discs. Many corollaries to Moore's law exist which suggest that progress will slow or hit a limit as fabrication techniques fail to keep pace with required performance improvements. Perhaps one of the most widespread is that the end of optical lithography is around the corner - at most two IC generations away.http://public.itrs.net ). This document is continually updated by the semiconductor industry and the latest version was published in December 2003. The table is taken from the roadmap and shows how the feature size of ICs will hopefully evolve.
While much of this piece is framed in the context of silicon ICs, it is worth noting that lithography has a much broader impact. Nanotechnology is emerging as a vibrant new field with enormous potential, and lithographic techniques, such as nanoimprint and molecular self-assembly, are likely to play a critical role in this new area of science.
The imaging performance of an optical-lithography system is described in terms of the Rayleigh resolution criterion, developed more than 100 years ago by Lord Rayleigh to describe the diffraction limits of astronomical telescopes. It describes the ability of an imaging system to resolve two closely spaced objects, and in the field of lithography effectively defines the smallest half-pitch (HP) of a feature on an IC.
It states that the critical dimension (known as CD and equivalent to the smallest HP that is possible) is equal to ~κ1λ/NA, where λ is the wavelength, NA is the numerical aperture of the imaging system and κ1 is a value between 0.25 and 1 that depends on the configuration of the illumination system and the optical response of the photoresist. A κ1 of 1 corresponds to an almost complete intensity null between two well resolved features. A κ1 of 0.6 corresponds to a ~20% intensity dip between two diffraction-limited spots for incoherent illumination conditions, and is the practical limit for conventional lithography. Various resolution-enhancement techniques (as discussed below) have been developed to extend κ1 toward the fundamental limit of 0.25.
Remarkable progress has been made in each direction. The wavelength of optical lithography used in chip manufacture has progressed from 432 nm (the G-line of Hg) to the current value of 193 nm (ArF excimer laser). Each shift has been accompanied by an extensive and expensive effort in developing appropriate optical sources and illumination imaging optics. Currently, further incremental improvements in wavelength do not appear feasible, largely because of the lack of suitable optical lens materials.
At the same time, there has been much improvement in the NAs of lithography lenses with values increasing from 0.1 to today's figure of around 0.9. Recently, immersion lithography - the addition of a liquid between the lens and the wafer - has been enthusiastically accepted as an important new technology. This technique increases the NA of the imaging system by a factor that is equal to the refractive index of the liquid and will lead to NAs greater than unity. NAs of ~1.3 are feasible with water immersion at 193 nm, and there is an active search under way for higher index fluids to increase the resolution further.
A final strategy for reducing the CD is decreasing the value of κ1 by using what are collectively known as resolution-enhancement techniques (RETs). However, decreasing κ1 also reduces the allowable variation in optical exposure across the image, so tighter process control is required.
Mask-based RETs include optical proximity correction, adding subresolution structures to the mask and introducing phase-shifts to turn the mask into a 3D object. Illumination-based RETs include off-axis illumination and imaging interferometric lithography. Combining all these techniques can produce κ1 values nearing the theoretical limit of 0.25.
With all of these improvements in NA, wavelength and κ1 , the imaging limit of optical lithography extends to an HP of 0.25 x 193/1.3 ~ 37 nm which satisfies at least the 45 nm HP technology node. If a sufficiently high index immersion fluid can be developed then perhaps this strategy can even satisfy the 32 nm HP node.
To go further will require further innovation. As noted earlier, the Rayleigh limit is on the pattern pitch, not on how small an individual feature can be printed.
Process nonlinearities, such as overexposure or oxygen plasma thinning (a process that decreases the dimensions of photoresist lines), allow further reduction of the CD to meet the needs of future lithography nodes. This is already common practice - and is the reason that the MPU CDs are smaller than the technology node in the ITRS.
Therefore one way to extend lithography is to print half of an image, for example with a 22 nm CD but on a 90 nm pitch (45 nm HP); store this image in a sacrificial layer on the wafer: print the second half in a second photoresist layer; and transfer a composite, interpolated image into the wafer. This is known as spatial-frequency doubling.
While this technique satisfies the resolution requirements, several problems are evident. First, it becomes increasingly difficult to maintain a uniform linewidth; this will require additional resolution enhancement as well as better process control. Accuracy of the overlay between the two patterns is an unresolved issue since the demands will be at the 1 nm level. Finally, this process demands at least two lithography exposures plus additional processing, and will be two or three times more expensive than conventional single-exposure lithography.
The end of optics has often been forecast, and the seers have always been wrong. It now appears that immersion lithography at 193 nm will extend to at least the 45 nm HP technology node (2010). With nonlinear extensions, there are no fundamental limits; however, enormous technical challenges remain. I am optimistic that these challenges will be met, and that optical lithography will be the technology of choice for volume manufacturing well into the future.