Incoming Inspection and Failure Analysis of CMP Consumables at the Semiconductor Fab

Dr. Norm Gitis and Michael Vinogradov

Center for Tribology, Campbell, CA 95008, www.cetr.com

Abstract

The variability in functional performance of polishing pads, slurries, conditioners and retaining rings (guides) is one of the major factors limiting the lot-to-lot uniformity of polishing and process yield. Both the yield and output of polished wafers are improved when the proper incoming inspection is put in place and no faulty consumables are used in planarization.

Performing incoming inspection on the production polishers is expensive and unrealistic. A bench-top controlled (with in-situ process monitoring) polisher mod. CP-4 provides an inspection solution, which is both time and cost effective. By polishing small (from 4” to 1”) wafer monitors on small (from 9” to 3”) pad coupons, utilizing the same process parameters (linear speed and pressure) and thus providing the same wafer removal rate as on the production polishers, it allows for fast screening of consumables.

A lot of experimental data on screening various pads, slurries and conditioners has been obtained for planarization of three types of wafers, with copper, tungsten and oxide films. The “good” (that is, within the process qualification window) CMP materials exhibited repeatable and reproducible CMP process characteristics, while the “bad” (that is, outside the process qualification window) ones exhibited different process parameters and so would produce out-of-spec wafer surfaces.

The incoming inspection and failure analysis of consumables represent the next step in how the major fabs utilize, optimize and maintain the CMP technology.

CMP Tribo-Metrology

Though chemical-mechanical planarization technology has been known for a quarter of century, its large-scale proliferation into the fabs has materialized only within the past several years. There are reputable predictions that in 2004 the semiconductor industry will have over twelve thousand polishing heads working in the fabs. The dramatic growth so far has been achieved by producing more and more polishers and consumables. Based upon earlier determined process windows, the industry has succeeded in the remarkable increase in number of wafers being planarized with currently acceptable yields. The further growth, however, and even the very existence of the CMP industry in the highly competitive environment where new technologies are developed all the time, will depend on the degree of progress it will achieve in terms of process reproducibility and yield.

Studies of surface finishing treatments have been a part of tribology, the science of friction and wear, or material removal. Surfaces of both a semiconductor wafer and polishing pad, together with polishing slurry between them, constitute a rather classical tribological system. This is sometimes referred to as a three-body interface, because it includes two solids in relative motion and the slurry. Modern CMP machines may have an abrasive pad conditioner rubbing the pad either during wafer polishing or between the polishing cycles, which makes the four-body picture more complicated. Any tribological system is characterized with classical parameters of coefficient of friction between the surfaces and their wear rate. A part of the later parameter has been widely used in the CMP industry as the removal rate of wafer material; the parameters of friction and pad wear have not become common yet.

The coefficient of friction (COF) is defined as a ratio of the tangential friction force, resisting relative motion of the surfaces, to the normal load pressing the surfaces together. In a case of high adhesion of wet smooth surfaces, the normal load shall be considered as a sum of the externally applied down-force and the internally developed adhesion (stiction) between surfaces. For simplification, the normal load is usually considered equal to the external down-force. To monitor friction, it is preferred to measure its coefficient instead of just measuring the force resisting the relative motion of rubbing surfaces. Indeed, there are cases when changes in down-force cause substantial changes in the friction force, while the coefficient of friction remains constant. Alternatively, sometime important changes in the coefficient of friction cannot be observed by monitoring the friction force due to periodic fluctuations of the down-force.

The total wear of the interface, typically measured in the direction perpendicular to the rubbing surfaces, consists of wafer and pad linear wear. The former one is the useful goal of the process,

the later one is a negative accompanying effect. In the wafer-pad interface, there is negligible pad linear wear (though there may be large pad glazing and deterioration) and supposedly substantial wafer material removal (up to the target level of wafer layer thickness). In the conditioner-pad interface there is a significant pad linear wear with no conditioner wear (except for unavoidable dulling and deterioration of a conditioner surface, as well as seldom loss of its abrasive particles). To monitor wear, it is preferred to measure the linear geometrical wear of each of the rubbing surfaces. As wafer material removal is of the order of nanometers, its measurements are very complicated during polishing, and are often performed after polishing. The pad wear is of the order of micrometers, and its measurements are straightforward.

Another important parameter of friction and wear is the acoustic emission (AE) from the contact of rubbing surfaces. Its spectrum may have numerous frequencies, corresponding to such different processes as plastic and elastic deformations of sub-surface material layers, micro-scratching and micro-fatigue, micro-corrosion and other electrochemical reactions, delamination of material layers. At given speed, higher frequencies reflect the processes on smaller micro-areas that spread into smaller depths. Thus, the mega-Hertz acoustics is more informative of the specific micro-tribo-processes on tiny micro-contacts, in comparison with a kilo-Hertz range reflective of integral characteristics of the interface and deci-Hertz range reflective of integral characteristics of the entire mechanical system.

Computerized real-time measurements and analysis of the coefficient of friction, contact high frequency acoustic emission and pad wear allow for the effective evaluation of dynamic characteristics of the polishing process, including rate and non-uniformity of material removal.

Indeed, it turns out that for the given combination of consumables and polishing conditions, the value of COF measured in the wafer-pad interface during polishing reflects the top layer on the wafer surface (Fig. 1). This allows for real-time, in-situ monitoring of the wafer surface without

optical observation. Either in-situ or post-polish analysis of the COF data allows for calculations of the time to remove wafer layer and so rate of material removal (though these calculations can take place only after the layer is removed). Moreover, as the transition from an upper layer to a lower layer does not happen instantaneously and takes some time, this time is a direct characteristics of the non-uniformity of material removal: the longer is the time of transition, the higher is the polishing non-uniformity.

Fig. 1. In-Situ CMP Monitoring On Copper/Low-K Wafers

Parallel AE measurements compliment the friction ones (see Fig. 1) in detecting the rate and end point of polishing. For example, in some cases the difference in friction of the upper and lower wafer layers may be insignificant, but their acoustic response to CMP may be sufficient to detect the transition from one layer to the other. Among the main benefits of the acoustic measurements in CMP are monitoring the intensity of polishing processes and detecting polishing conditions when wafer layers, for example low-K polymers, delaminate (Fig. 2).

As CMP volumes are growing and the industry is maturing, Fab engineers are facing the pivotal issue of process repeatability and reproducibility. Modern polishers have very good control of motions and pressures, incoming wafers are typically not too different, so the main causes of wafer-to-wafer nonuniformity are lot-to-lot fluctuations in functional properties of consumables.

Fig. 2. Effect Of Down-Pressure On Copper/Low-K Delamination

Fig. 3. Bench-top CMP Machine

The novel bench-top polisher has a small footprint (3 by 2 feet), requires no special facilities, can be easily installed anywhere in the Fab. It provides a fully controlled, instrumented CMP process on 2 “ - 4” (for mod. CP-4) or 1” - 2” (for mod. CP-2) wafers or wafer coupons and 4” – 9” (for mod. CP-4) or 3” - 6” (for CP-2) pad coupons. Wafer and pad coupon cutters are supplied with the machine. The tester has precision translational (for wafer and conditioning disc) and rotational (active for pad and wafer, passive for conditioning disc) motions, with programmable speeds from 1 to 500 rpm. A down-force is applied via a closed-loop servomechanism and can be either kept constant or changing in the range 1 - 500 N. Static and dynamic friction force, torque and coefficient, contact high-frequency acoustic emission and pad wear depth can all be measured in-situ at a total sampling rate of 20 kHz, displayed and recorded for further analysis.

Slurry Inspection

An example of utilizing the CMP Tester for testing polishing slurries is illustrated in Fig. 4. A semiconductor manufacturer has accumulated 7 batches of supposedly the same common slurry for tungsten polishing, but it had suspicions that 2 of them had deteriorated. Several expensive multi-wafer tests on production Mirra polishers have neither confirmed nor disproved the suspicions. For a quantitative fast estimate of their functional properties, the slurries were then tested on the CMP Tester, with the same process parameters (linear speeds, pressure and removal rate) as on the Mirra, with continuous acquisition of the frictional and acoustic signals that allow for determination of three important functional parameters, namely: time of complete tungsten removal, time of complete under-layer removal, and intensity of polishing (characterized by the level of acoustic signal).

It turned out that six of the tested slurry batches removed both the tungsten and under-layer before the maximum times allocated on it, while one batch (batch 1, blue lines) was unacceptable due to both longer time for material removal (the under-layer did not get fully removed during the test duration) and lower level of polishing intensity. The other batch under suspicion (batch 2, red lines), though exhibited slightly rougher acoustic profile and slightly different rate of tungsten removal, still fit the three criteria and so got accepted for production.

Another example of slurry pre-screening is presented in the next-page table. 3” copper test wafers were polished in the same conditions as the 8” product wafers on the corresponding Mirra polisher, namely: pressure 4 psi, speed 120 rpm, duration 1 min (after 0.5 min of ex-situ conditioning), removal rate about 6,000 A/min. Three batches of the common commercial copper slurry were inspected, lot 1 (in spec), lots 2 and 3 (under suspicion).  The bench-top results showed the good lot 1 to have COF of 0.60, AE of 1.34, RR of 6,100 A/min, the faulty lot 2 to have the same friction and removal rate but higher AE of 1.71 (indicating scratching resulting in wafer defects), the faulty lot 3 to have lower COF of 0.53 and RR of 5,500 A/min (lower rate of polishing). Verification of these results on the production polisher fully confirmed the screening data, with excellent process correlation between the production and benchtop machines.

Fig. 4. Quality Control Testing of Tungsten Slurry

Table. Screening of 3 Lots of Copper CMP Slurry

Pad Inspection

A number of tests can be done on the bench-top CMP Tester with polishing pads. Deformation tests (with an upper indenter of either flat or spherical shape, moving up and down, no wafer, on the stationary pad) are used to evaluate elastic, plastic and creeping properties of either complete pads or their working (polyurethane) and supporting (foam) layers separately. Wear tests (with an upper conditioner either rotating or stationary, no wafer, on the rotating pad) are used to characterize pad wear resistance. Scratch tests (with a sharp upper indenter moving both vertically and horizontally on the stationary pad) are used to measure pad scratch resistance. Functional polishing tests (with either in-situ or ex-situ conditioning) allow for the quantitative estimate of pad functional properties, by determination of four important functional parameters, namely: time of wafer layer removal (characterizing the removal rate), transition time from one wafer layer to another (characterizing the polishing non-uniformity), acoustic level (intensity of polishing) and pad wear (durability). 

Fig. 5. In-situ Pad Friction and AE Measurements

Fig. 6. In-situ Pad Wear Measuremnts

Plots in Fig-s 5 and 6 show tungsten polishing results for two pads, pad 1 with more intensive polishing and faster tungsten removal (Fig. 5) and less pad wear (Fig. 6) versus pad 2 with slower tungsten removal (Fig. 5) and higher pad wear (Fig. 6).

1. Tribo-metrology is one of the key technologies for the control and optimization of CMP materials.