Tungsten Chemical Mechanical Polishing Endpoint Detection
- PDF / 737,980 Bytes
- 6 Pages / 417.6 x 639 pts Page_size
- 103 Downloads / 232 Views
rates. Historical CMP recipes have been based on time with separate recipes for contact and via layers. The Luxtron Real-Time Controller (LRTC) system provided the capability to use endpoint methodology to manage end-of-process versus fixed time recipes. The system controller is designed to process and analyze signals that are relevant to the change in the tungsten CMP process and equipment. Implementation of active analog gain, antialliance and anti-harmonic filtering in combination with higher than 12 bit analog to digital data acquisition methodology enhanced the signal to noise ratio. Furthermore, using an Automatic Gain Control (AGC) and signal normalization, allowed optimization to yield improved signal differential detection. Numerous combinations of signal averaging to smooth out the incoming motor current noises were evaluated. Initial endpoint recipe development did not result in good repeatability. The recipe would endpoint on several wafers, with the remainder defaulting past the endpoint set time. During the interpretation of the curve traces it was determined that a portion of the trace behavior was characteristic of the polisher and not the wafer film material. In principle, the endpoint methodology considers that as a particular layer clears, the material on the next layer exhibits a different coefficient of friction (change in the current due to a difference of frictional loading). However another factor affecting the curve traces was found to be the carrier arm movement motion across the primary platen and the conditioning head (both 109 Mat. Res. Soc. Symp. Proc. Vol. 566 ©2000 Materials Research Society
move back and forth across the radius of the polishing pad). The load that the carrier itself and the conditioning head impose on the primary platen was found to have an effect on the incoming signal, resulting in an oscillation of the curve traces. It was therefore necessary to remove the effects of these oscillations through the application of data averaging. This modification allowed improved synchronization with the periodic behavior of the carrier head and the conditioning transverse motor. Various experiments were carried-out with this modification relative to the detection criteria. Extensive off-line reprocessing of passive data was evaluated in order to optimize the endpoint recipe prior to final testing on production material. During a three month period the following was evaluated: 1. Repeatability of the endpoint process, 2. Endpoint methodology versus CMP process and 3. Electrical analysis on device material. EXPERIMENT The traces below illustrate unconditioned and conditioned data with endpoint. The trace as shown in Figure 1 represents the raw current signal from the motor. By applying gain control and averaging to the raw data, a conditioned trace results as shown in Figure 2. Both single and double peaks were observed on given material. In order to endpoint on the second peak, system parameters were defined to look for endpoint after a designated amount of time had transpired (
Data Loading...
