PALASH DAS and KEN REBITZ
Mercury arc lamps have been the primary lithographic illumination source for integrated-circuit (IC) manufacturing for more than 30 years. For several reasons, these lamps are not optimal for manufacturing future generations of semiconductor chips, which will require critical feature sizes smaller than 0.25 µm. Mercury arc lamps emit a wide spectrum of light, which degrades resolution and limits their effectiveness for critical feature sizes. Also, because only a small portion of mercury arc lamp emission is in the deep ultraviolet (DUV) spectrum, the small amount of DUV light delivered to the surface of a wafer necessitates longer exposure times and slows production throughput. Furthermore, the high-numerical-aperture (NA) lenses on new refractive tools limit depth of focus, and this reduces the bandwidth of light allowed to less than that provided by mercury arc lamps in sufficient quantity for production.
Semiconductor manufacturers are currently gearing up for production of next-generation devices, and the shift from mercury arc lamp-based lithography tools to excimer laser-based DUV lithography tools has begun in earnest. Krypton fluoride (KrF) laser-based lithography systems are currently available from all suppliers and satisfy the optical requirements for pilot and next-generation DRAM and logic production. The excimer laser`s transition from research and development into production, however, has increased demands by chipmakers for information about laser manufacturability, uptime, and cost of ownership. Chipmakers now routinely require reliability, availability, and maintainability (RAM) data from laser manufacturers to estimate uptime and operating costs.
Advanced lithography tools fall into two categories: refractive lens reduction and catadioptric (reflective). In refractive-lens-reduction steppers and scanners, a reticle is imaged onto a wafer through the lens—this system requires a narrow-bandwidth (<3 pm, 95% energy) source to maintain depth of focus. A catadioptric scanning system images the reticle pattern on the wafer by scanning a slit exposure across the reticle using both refractive and reflective elements to produce an image on the wafer. This system allows the spectral bandwidth of light to be as broad as 150 pm or even more.
Current UV-laser lithography tools are based on KrF excimer lasers emitting at 248 nm. Next-generation argon fluoride (ArF) lasers emitting at 193 nm have received substantial support from leading semiconductor industry organizations, including Sematech (Austin, TX), and this technology is expected to be in production plants as early as the year 2000 for manufacturing chips with critical geometries smaller than 0.18 µm.
Krypton fluoride lasers for lithography
Although its output is spatially incoherent compared to other lasers, the KrF excimer laser is spectrally narrower than a lamp, which allows for a simpler optic and illuminator design. The laser is relatively efficient, typically 0.6% for a spectrally narrowed system and more than 2% for a broadband laser. The high-brightness output of the KrF laser facilitates efficient coupling of laser energy into the stepper or scanner, which reduces wafer exposure time and increases throughput.
In addition, a KrF laser provides precise illumination on demand with no reduction in output power and no consequent loss of production throughput, which is the case with a mercury arc lamp near the end of its life. The KrF laser is suited to efficient implementation of enhanced illumination schemes—such as quadrapole or annular—with minimal loss of throughput. And remote installation of the laser eliminates unnecessary thermal loading on the stepper or its enclosure.
The relationship between a KrF laser and a stepper or scanner is multifaceted (see Table 1). The linewidth of a free-running KrF excimer laser is approximately 300 pm, so a line-narrowing module "compresses" the DUV emission suitable for sharp focus for either a refractive or reflective projection tool. The spectral bandwidth and spectral energy distribution—that is, the energy within the spectral region around the center line—play an important role in the lithography process. Depending on the type of projection optics, the spectral distribution may not exceed a few picometers for refractive optics or a couple hundred picometers for reflective (catadioptric) optics.
A high-NA (>0.5) catadioptric system can use much of the broadband spectral output of the laser. Due to the high-gain, short-duration nature of the KrF laser pulse, the design of the line-narrowing module is not trivial, especially for refractive tools. The required spectral distribution is achieved by controlling the evolution of the laser pulse and by maximizing the single-pass efficiency of the line-narrowing module.
Reliability and cost of ownership
The cost of an excimer laser, although higher than that of a mercury arc lamp system, has come down considerably since its introduction into photolithography eight years ago and is now a fairly small contributor to the overall cost of operation of a photolithography system. New developments have improved laser reliability and reduced cost of operation, making the KrF excimer laser well suited for mass-production environments.
The RAM data required by chipmakers to integrate a laser system into high-volume manufacturing plants typically take several years to accumulate. Due to rapid industry acceptance of the KrF laser, however, RAM data on the laser are needed more quickly. The general approach followed by laser manufacturers in determining RAM data has been to lifetest one laser for a certain duration and document the failure mechanisms. Cost of ownership is then estimated from the cost of replacement parts and subsystems. However, nuisance failures such as a blown fuse, a defective valve, or software bugs are rarely addressed, and these failures alone can cause significant downtime during the manufacturing process. Therefore, RAM data are merely an educated guess by the laser manufacturer.
To provide more accurate and meaningful operating costs and uptime data, Cymer has adopted Sematech`s failure reporting analysis and corrective action system (FRACAS) to determine the manufacturability, reliability, and uptime performance of the KrF excimer laser. The FRACAS process is a closed-loop method developed to record, group, and analyze failures and determine preventive maintenance actions.
It provides the required data for corrective action, highlights developing failure modes, contributes data for statistical analysis, and can be used to ensure closure of all problems. Over time, the use of FRACAS results in a reduction of failure events, ultimately leading to improved reliability and manufacturability. Using FRACAS, RAM data can be accurately calculated.
In a recent ten-week test, a Cymer Model ELS-5000 laser was operated continuously in stepper mode. Test results showed that the system`s mean time between failures was greater than 1000 h and the mean time to repair was approximately 2.6 h. Most notably, greater than 99% uptime was attained. The emphasis on replacement parts and subsystems has resulted in more stringent reliability requirements for the subsystems—not just for the laser.
In addition, recent developments have significantly increased the lifetimes of the discharge chamber and the pulsed power module of the laser. The increase in the lifetime of the chamber is attributable to the use of optimized fluorine-compatible chamber materials. The improvements in the solid-state pulsed power module are largely traceable to the replacement of the thyratron switch traditionally used in excimer laser designs with a solid-state switch. Although the switch is more complex, the advantages of the solid-state pulsed power module over the thyratron more than compensate for the complexity.
These advantages include elimination of a costly and short-lived thyratron switch. By selecting suitable design parameters, it is possible to operate the switch in the pulsed power module in a manner that provides virtually infinite lifetimes. Furthermore, the occasional misfires or prefires (approximately one per million pulses) associated with thyratron discharges are eliminated by silicon-controlled rectifier (SCR) switching—an effect that can be important for scanning applications, in which a single bad laser pulse can result in the loss of an entire chip.
Next-generation excimer lasers
Based on an ArF laser with an output wavelength of 193 nm, the next generation of excimer lasers will be capable of imaging critical features as small as, or even smaller than, 0.12 µm. The principles of operating an ArF laser are similar to those of a KrF laser, although ArF lasers still lag behind in performance parameters such as repetition rate, optical losses, and conversion efficiency. Still, semiconductor industry experts predict a smooth transition between the two technologies when the time comes. Production-worthy ArF lasers are expected to be available in 1998.