Abstract
The PadProbe® with simultaneous
real-time measurements of PadFriction and PadWear was used to evaluate a copper CMP process and
consumables.
During conditioning, PadFriction
reaches its optimum level. Any further continued conditioning has to be
eliminated, as it wears the pad without any improvement to its surface. The
moment when PadFriction is stabilized at its optimum
level defines the end-point of conditioning.
Monitoring PadWear allows for
determination of pad life and can trigger an alarm signal when the critical
wear level is reached, for example, based on the depth of pad grooves.
When PadFriction drops out of the
process window, there are two common causes: if PadWear
is close to its critical level the pad has to be replaced, if PadWear is still low the conditioner has to be replaced.
The observed correlation of the in-situ measured PadFriction with post-polish measured Removal Rate and
Non-Uniformity of Polishing confirms the PadProbe®
usefulness for in-situ process control.
Several types of conditioners were tested on two types of
pads. The simultaneous measurements of PadWear and PadFriction allowed us to effectively choose the optimum
conditioner-pad combination.
Keywords: CMP, PadProbe, pad
friction, pad wear.
Introduction
In semiconductor manufacturing, chemical mechanical planarization (CMP) is one of the least stable processes.
Its performance stability is affected by a large number of variables, including
process parameters and consumables. CMP is also one of the most expensive
processes, with most of the cost of ownership in the CMP consumables, such as
slurry, polishing pad and pad conditioner. In the industry practice, CMP
process and consumables evaluation, as well as CMP process control, rely
heavily on feedback from metrology measurement of polished wafers. In many
cases, it means waiting for post-polish wafer electrical and defectivity tests before a problem be
addressed on a polishing tool. Thus, it is widely desired that in-situ
monitoring of CMP process can be established. By using the PadProbe®,
the real-time pad friction can be monitored, which provides information on pad
surface properties and slurry flow dynamics on the pad, which are directly
related to polishing performance. In addition, by monitoring pad wear rate, the
end-of-life for both pad and conditioner can be readily determined.
Experimental Results
The PadProbe® was installed on the
CMP machine made by Ebara Corporation.
PadWear was measured during
polishing and conditioning. The results, presented in Fig. 1, show measurable
wear during ex-situ conditioning and no wear during pad polishing. Monitoring
pad wear allows the process operator to determine when it reaches its critical
level, and so to know exactly when to replace the pad.
|
|
|
Fig. 1. PadWear Dynamics During Polishing
and Conditioning |
PadFriction was measured during pad conditioning. The results are presented in Fig. 2 for two similar pads. Out of 25 wafers polished on each pad, three (2nd, 13th and 25th) were blanket monitor-wafers that are presented in the plots. During conditioning PadFriction changed in a quite repeatable way, reducing from about 0.4 to about 0.2 and then stabilizing. The exceptions were blanket wafers #2 that were started on a preconditioned pad and so had the 0.2 level of friction coefficient from the beginning. As the friction stabilization took place within less than 20 s, the rest of the 45-s conditioning time was not only wasted in terms of polishing productivity, but also produced excessive pad wear. Thus, use of the PadProbe® to monitor the end-point of conditioning and to stop conditioning when PadFriction is saturated may dramatically reduce pad consumption.
|
|
|
Fig.
2. PadFriction During
Conditioning |
PadFriction
was measured during polishing of the blanket wafers 2, 13 and 25, while product
wafers were polished in between. Four series of polishing tests were conducted,
with and without ex-situ conditioning, each on two pads. The PadFriction plots for each of the four series are presented
in Fig. 3. The data is summarized in the next page table.
- without conditioning, both pads exhibited substantial
degradation in removal rate RR and within-wafer-nonuniformity
WIWNU, which was well reflected in PadFriction,
- with ex-situ conditioning, both pads had stable process
characteristics of RR and WIWNU, reflected in the stable PadFriction,
- there is a correlation between PadFriction
(measured at the end of conditioning or beginning of polishing) and common
process parameters of RR and WIWNU.
|
|
|
Fig. 3. Polishing Tests With and Without Conditioning |
