BILL FRICKE, MIKE GANOPOULOS, and CLIFFORD MARTIN
The microlithography world continues its push toward smaller linewidths for computer chips at a speed that follows Moore's law, according to which the number of transistors that can fit on an integrated-circuit (IC) chip doubles every 18 months. This rapid pace has led the lithography community to attempt to extend current optical-lithography techniques down to 157 nm. Such a technological push must include the development of better detectors to monitor and control processes.
Choosing a probe for 157 nm
Five beam parameters are required to establish the specifications of an optical lithographic probe: the wavelength (in this case 157 nm), the size of the beam (either diameter or height and width), the pulse-repetition rate, and the maximum and minimum pulse energy. The beam size and maximum pulse energy yield the fluence, important for pulsed-laser damage and saturation issues. The pulse-repetition rate and maximum pulse energy yield the average power, important for damage thresholds related to the average heat load. The beam size by itself determines the minimum input aperture of the probe. The pulse-repetition rate yields the maximum response time of the probe. The maximum pulse energy gives the upper limit of the dynamic range.
Stability and drift—both reversible and irreversible—are concerns for a lithographic probe. Compensation for reversible changes, such as those due to temperature, is often possible. Temperature drift can be canceled by monitoring the temperature of the probe during use and applying an empirically derived functional correction to the output signal.
Irreversible changes, such as ablation of the absorber material, are not only more difficult to measure, but also harder to model. They are more difficult to measure because each sample is damaged as part of the measurement, so many samples are required to provide good degradation statistics. They are more difficult to model because there are many independent parameters that must be accounted for, such as beam location and size, fluence, and repetition rate.
Although absolute calibration of a probe is important, it is not always required. An uncalibrated stable probe is more useful than a calibrated unstable probe. This is especially true for high-precision microlithography, because it is usually more important to be able to set exposure parameters to the same values day after day than to know the total dose accurately. Calibration and calorimetry at 157 nm require great care.
Although linearity of output versus input simplifies data analysis, it is not required for many uses. Thermocouples, for example, are very useful nonlinear devices for which a software look-up table can be used to find the temperature when given the output voltage.
Probes types currently available for deep-ultraviolet (DUV) applications include semiconductor (silicon and gallium nitride), thermal (thermopiles and pyroelectric), and photomultiplier tubes. In addition, a probe that is a relative newcomer to DUV detection is based on the UV downconversion process developed and patented by Star Tech Instruments (Danbury, CT).
The semiconductor option
In semiconductor detectors, the bandgap of the semiconductor material determines the long-wavelength detection limit. Generally, at wavelengths longer than the bandgap, the material is transparent and no electron-hole pairs are produced. At short wavelengths, the absorption depth becomes very small, and the depletion layer must be made very thin. As a result, few electron-hole pairs are formed in the depletion layer.
The bandgap of silicon is 1.12 eV, corresponding to a long-wavelength cutoff of 1100 nm. Although the practical short-wavelength limit for silicon is around 300 nm, it can be used at wavelengths as short as 190 nm. Unfortunately, silicon detectors are easily damaged by UV radiation. Degradation begins immediately and causes large changes in the photodiode responsivity after only a few hours, making them unusable for precision measurements in the UV.
The bandgap of gallium nitride is 3.18 eV, corresponding to a long-wavelength cutoff of 387 nm. Gallium nitride has little response at longer wavelengths, making it useful when solar-blind detection is necessary. Its short wavelength limit is 180 nm. Diamond UV photodetectors have a useful spectral range from 180 to 225 nm.
Thermal sensing of light
Thermal detectors rely on the temperature rise of the detector material to produce a current flow or charge that is proportional to the incident power or energy. Blackened absorbing surfaces enhance the sensitivity and provide a broad, flat spectral response, while reflective coatings raise the damage threshold.
A pyroelectric crystal produces a charge polarization between the front and back surfaces of the crystal in response to a temperature change inside the crystal. The total charge is proportional to the average temperature change of the bulk material. Pyroelectric detectors are only suitable for detecting pulsed sources. Large-area pyroelectric detectors are generally limited to low repetition rates (a few hundred hertz). Their spectral response is determined by the front surface coating. Although many black coatings should absorb at 157 nm, there is little data supporting their use at wavelengths shorter than 190 nm. This is largely a calibration issue at this time. Damage thresholds are typically 400 mJ/cm2.
A thermopile consists of an absorbent element—usually a blackened metal disk—with thermocouples attached to the back of the disk and connected in series. Uniform spatial response of the device requires careful placement of as many thermocouples as is practical. Response time is determined by the time that it takes the disk to thermally stabilize. Fast thermopiles must be very small. The same issues that limit the use of pyroelectric detectors at 157 nm also limit the use of thermopiles at 157 nm.
Photocathodes and photoelectric devices
A basic photomultiplier consists of three components. A photocathode produces electrons when it absorbs a photon. Behind the photocathode is an evacuated region containing a series of dynodes that produce many electrons for every one produced by the photocathode. Finally, an anode after the last dynode collects the cloud of electrons, which become the output signal.
Because of the huge amplification factor, photomultipliers are extremely sensitive and, with care, can count single photons. The high sensitivity of photocathodes limits their use to low-energy sources. The photocathode material determines the spectral range.
Optical downconversion
A probe based on the downconversion technique eliminates many of the problems associated with other types of DUV detectors and at the same time can be made to sense light down to x-ray wavelengths. In such a probe, absorbed DUV photons produce visible fluorescence in certain kinds of crystals. The fluorescence is linearly proportional to the incident pulse energy or continuous-wave power. The visible fluorescence can be easily detected using inexpensive silicon photodiodes. The input aperture of the probes is limited only by the size of available crystals. The response time is determined by the fluorescent lifetime of the crystal, not its size, so that response time is independent of input aperture.
A 100-mm-diameter downconversion probe can detect 10-kHz pulse trains with no compromise on sensitivity. Such probes are insensitive to out-of-band (visible and longer) radiation because the materials used do not produce fluorescence when irradiated in the visible. Damage thresholds are high, typically exceeding 500 mJ/cm2. Long-term responsivity drift is caused by the temperature rise of the probe and is reversible, with a magnitude of less than 5% for standard probe designs. Probes designed for precision measurements in the DUV have long-term drift of less than 1%. Temperature compensation can reduce this drift even more.
Another advantage of downconversion probes is their design flexibility. Because the emitted light can be transmitted to detection modules via optical fibers, the size and shape of the detection area is not restricted by detector limitations. In addition, the detection modules can be placed far from any source of electromagnetic interference.
Energetic photons can cause damage
Damage to the sensitive element in a 157-nm probe is an issue even at low average power or fluence. This is because the photon energy at 157 nm is high enough to remove material from surfaces even at room temperature. In general, if the beam produces an audible snap when it hits a surface, some damage has occurred. If the surface is the sensitive element of the probe, then a small, irreversible responsivity change has taken place. Over time and many exposures, the small change will become noticeable and the probe will have to be recalibrated and, eventually, replaced.
Deep-ultraviolet radiation can also cause color-center formation in optical materials. Although color centers are an important consideration for any UV probe with a window, they are especially important for downconversion-type probes because they rely on the detection of visible light. Because the visible light must travel through a portion of the crystal prior to detection, its responsivity may be affected by color centers. Probe materials developed at our company have been selected to be immune to color-center formation.
Novel uses
Because the fluorescent light-intensity distribution in a downconversion crystal is an accurate representation of the energy distribution of the incident UV beam, accurate imaging systems for x-ray through DUV sources are possible that use conventional low-cost optics and charge-coupled-device cameras (see figure).
Calorimeters for 157 nm are a class of "detectors" all by themselves. Our company is developing a calorimeter based on the electrical substitution method that has been designed to operate at all DUV wavelengths, including 157 nm. The sensitive element is enclosed in a chamber that can be evacuated or backfilled to reduce errors due to transmission losses through the air. The computed measurement error of the device, including the calorimeter constant, the absorption of the sensitive element, and the transmittance of the calorimeter window, is less than 3%. It is expected that the measurement error will be less than 1% once the setup and calibration procedures have been refined.
Bill Fricke is president and Mike Ganopoulos is an engineer at Star Tech Instruments Inc., POB 2536, Danbury, CT 06813; e-mail: [email protected]; Clifford Martin is a partner at Odyssey Consortium, 304 Main St., Perry Plaza PMB-395, Norwalk, CT 06851: e-mail: [email protected].