Temperature control slows RTPs advance
12/01/1996
Temperature control slows RTP`s advance
According to speakers at the recent 4th International Conference on Advanced Thermal Processing in Boise, ID, temperature measurement and control are among the most pressing and perplexing issues facing designers and users of rapid thermal processing (RTP) systems. The SIA National Technology Roadmap calls for ?5% gate oxide thickness control for the 0.25-?m device generation, which will require ?5?C temperature control (3 s). While thermal budget demands suggest that RTP will be used for many of these advanced processes, temperature control is substantially more difficult than in batch systems.
 |
Heat transfer in a typical RTP system. The I1 to I8 are radiant energy fluxes, while ILT and ILB are radiation from the top and bottom lamp banks. The e are emissivities in the long and short wavelength bands, while the W are black body radiations
In batch systems, the wafer is immersed in a stable thermal field which is gradually raised to the process temperature. The wafer is in thermal equilibrium with its surroundings, so temperature can be measured anywhere within the thermal field. In RTP, on the other hand, the wafer is heated and cooled very quickly - ramp rates approach 50?C/sec - and remains at the process temperature for only a few seconds. The wafer is generally not in thermal equilibrium with the furnace, so its temperature must be measured and controlled directly.
At present, all single-wafer RTP systems measure substrate temperature by pyrometry, typically by detecting the radiation emitted from the back side of the heated wafer. A detailed sensitivity analysis, presented by Muhammed Refai of Integrated Systems, showed that emissivity variations in the system can have a substantial effect on the wafer temperature. The emissivities with the most impact are, in order, the chamber top wall, the short wavelength (<4 ?m) wafer bottom, the bottom wall, the short wavelength wafer top, and the long wavelength (>4 ?m) wafer bottom. (RTP analyses typically use this two-band wavelength model because quartz is transparent at wavelengths below 4 ?m, and opaque at longer wavelengths.) Of these, the wafer emissivities are the most difficult to predict or control. Emissivity varies with coating thickness and composition, as well as with wafer doping. G. Chen, of Duke University, showed that the optical properties of silicon may fundamentally limit the accuracy of optical pyrometry: the lamps commonly used for wafer heating excite free carriers in silicon. These nonequilibrium carriers alter the emissivity, lowering the measured temperature relative to the actual temperature. This photothermal effect, as large as 40% in lightly doped wafers at low temperatures, became very small when the sample temperature exceeded 1000K. In production systems, a variety of methods are typically used to either estimate or compensate for emissivity variations.
Still, as conference co-chair Bohumil Lojek pointed out, all of these methods assume that the temperature of the backside (wafer level) is the same as that of the active device regions on the front side (die level). Lojek showed that the wafer level temperature may be as much as 4? less than the die level temperature. This behavior occurs for two reasons. First, due to interface scattering, the thin films on the wafer surface will have thermal conductivity as much as two orders of magnitude lower than comparable bulk materials. Second, the substrate`s absorption and reflection of IR radiation will depend on doping and layer composition.
Paul Timans, research and development manager at AG Associates, discussed four important alternatives to emissivity-corrected optical pyrometry: open loop intensity control (OLIC), use of a radiant hot plate, use of a backside shield, and direct thermocouple control. OLIC processes the wafer through a preprogrammed lamp intensity cycle. The temperature is calculated from the heat fluxes in a simplified model (see figure on page 34). The model treats each element in the system - lamps, quartz isolation tube, and the wafer itself - as both an absorber and an emitter of radiation. The steady-state temperature of the wafer and the thermal response of the quartz plates in the OLIC model will both vary dramatically with wafer front and backside coatings. Once calibrated to particular surface coatings, however, the model relies on the integrated emissivity across the wafer surface, and is thus relatively insensitive to film uniformity variations.
So-called "hot plate" systems have also been used to reduce the effects of backside emissivity variations. The wafer is placed above a radiant plate whose temperature is measured by pyrometry. The wafer is heated by a combination of radiation and conduction through the ambient gas, if any. Timans` analysis found that lightly doped wafers, which absorb very little of the plate`s IR radiation, take about 200 sec to reach the steady state in vacuum. Addition of a nitrogen ambient allows for much faster heating, since both conduction and radiation components are present, but the heating behavior strongly depends on the mechanical gap between the wafer and the hot plate. The analysis also found substantial variations in steady state temperature, depending on coating. The largest variation, 53?C, was observed between top-coated and back-coated wafers in vacuum. According to Timans, "these results suggest that emissivity independence cannot be achieved on product wafers."
The third alternative uses an opaque shield between the wafer and a bottom bank of lamps. If the bottom surface of the shield and the top surface of the wafer are optically identical and the top and bottom lamps deliver equal power, then the steady state temperature of the shield and wafer are always identical. If either of these surfaces changes - as would be likely in a production environment - the two temperatures will no longer be equal.
Since none of these optical methods was able to achieve full emissivity independence, Timans suggested that future temperature control methods may require entirely different physical principles. One alternative, direct thermocouple control, is simple and reliable, and has relatively low cost. A thermocouple inside a thin SiC sheath can simply be placed in direct contact with the wafer. However, as Lojek pointed out, thermocouples have limited lifetimes at these temperatures. Mounting the thermocouple so as to minimize contact resistance and avoid perturbing the thermal field is an important practical challenge for this approach. - K.D.