Semiconductor Wafer Lapping and Displacement Measurement

Description

Semiconductor Wafer Lapping and Displacement Measurement

This application note explains how MTI’s Accumeasure technology was used with a lapping machine to measure displacement (wafer material removal) and determine the new semiconductor wafer thickness. Changes in electrical capacitance (displacement) were measured and then directly converted into a 24-bit digital reading to obtain precise digital thickness measurements.

During lapping, a wafer of known start thickness is placed on a rotary lapping table. The backside of the wafer faces downward and toward a lapping surface that rotates and removes unwanted material. The amount of material that is removed varies by device type, and the entire semiconductor fabrication process is tightly controlled to avoid removing too much material and ruining the wafer.

Lapping Displacement

In this application, a semiconductor wafer was positioned between a Pyrex plate and fluid barrier directly above a lapping wheel. A heavy weight is placed on top of the wafer to apply constant pressure to the wafer see Fig 1. A Capacitive probe faces the weight which is free to rotate in a guide ring. It’s important to note this weight is ungrounded but MTI’s push pull probes can monitor the displacement even with ungrounded targets. The fluid acts as a lubricant between the work material and the abrasive grains. As the wafer abrades (thins) the weight drops down due to gravity and the wafer thickness is proportional to this difference in height (as measured from the wafer initial thickness).

Basic lapping setup with one of two MTI capacitance probes shown

Figure 1 – Basic Lapping Setup with one of two MTI capacitance probes shown

Probe mounting fixtures were fastened to the rotary table and a capacitance probe was attached to an arm that was positioned above but did not make contact with the top of the metal weight. A second capacitance probe (see Fig 2) was installed to monitor the lapping plate displacement. The change in the average difference between the weight and the lapping plate is equal to the material removed from the wafer. Signals from MTI capacitance probes were sent to an MTI amplifier that was connected to a computer for lapping thickness control.

Lapping setup with full MTI Accumeasure system, including both capacitance probes.

Figure 2: Lapping setup with full MTI Accumeasure system, including both capacitance probes.

MTI’s push-pull capacitance probes were used to determine changes in height for the metal weight and were chosen because the lapping weights could not be grounded. These non-contact probes have a 2 mm gap range and 0.2 µm accuracy. By comparing height changes to a preset limit, it was possible to track apparent decreases in semiconductor wafer thickness. Alarms could be generated for stop conditions and a master network computer was used for master control.

MTI push-pull capacitance probe with push-pull amplifier

Figure 3: MTI push-pull capacitance probe with push-pull  amplifier

Push-pull probes feature two capacitance sensors built into one probe body. Each sensor is driven at the same AC voltage with a 180 degree phase shift between signals. The shift allows the current to travel across the target surface rather than through the target to ground, eliminating inaccuracies created by poorly grounded targets.

Measurement Concept

This experiment involved signal processing both on the MTI Accumeasure and on the host computer using LabVIEW software.

Signal processing on the Accumeasure for the first probe

  • Anti-aliasing filter: 5kHz cut-off frequency
  • Sample frequency: 20kSPS
  • Additional smoothing was done in the LabVIEW program as needed

Figures 4 and 5 show the signal from the top of the weight (time series).

Expanded signal from the top of the weight

Figure 4: Signal from top of weight expanded

Signal from top of weight compressed (multiple revolution).

Figure 5: Signal from top of weight compressed (multiple revolution). The median average of the waveform was computing using LabVIEW. In this figure, it would be equivalent to “1”.

Signal processing on the Accumeasure for the second probe

  • Anti-aliasing filter: 5kHz cut-off frequency
  • Sample frequency: 20kSPS
  • Additional smoothing was also done in the LabVIEW program as needed to remove the groove signature
Signal from the lapping plate push-pull probe

Figure 6: Signal from the lapping plate push-pull probe. The positive spikes are caused by the grooves in the lapping plate and must be ignored because they upward-bias the average displacement.

The LabVIEW software was used to complete the mathematical calculation and determine the total wafer thinning based on the original wafer thickness.

  • To determine the wafer material loss, we measured the median surface displacement of the lapping weight and the median surface of the lapping disc with the lapping disc grooves filtered-out. The median surface calculation and groove filtering were both done in LabVIEW. We were able to ignore the grooves by computing the median of the signal with a set number of samples.
  • We auto-zeroed the difference between the two surfaces (in LabView) just as the lapping process began. To determine the median surfaces, it is necessary to have at least a few rotations. Essentially, this is the RMS displacement value.
  • As the semiconductor wafer became thinner, the difference between the two median surfaces was equal to the amount that the wafer was thinned.
  • The customer (Computer) observed the material loss to determine the new wafer thickness. The difference between the auto-zero start point and the negative thickness change equals the amount of material removed. The original thickness of the semiconductor wafer minus the material removed equals the new average wafer thickness. A stop lapping alarm was set to the amount of material that needed to be removed.
Software processing sequence to compute wafer thickness

Figure 7: This is the software processing sequence to compute the wafer thickness. Spike removal was accomplished by computing the median of the signal. (Boxcar averaging with a set number of samples.)