|
|
 |
Monitoring copper removal rate, or end point detection in CMP processes, is a critical element in any CMP tool. Almost every CMP tool
manufacturer faces this problem. Some use an optical technique to detect end points, some use an eddy current technique to measure removal rates, but none of the CMP tool manufacturers are capable of delivering a
completely controllable CMP system to the end user. As a result, some find a different direction to avoid CMP processes due to high stress on low K materials or attempt to use a different pad, a different slurry, a
different pressure, a different tool design, etc. Regardless of the CMP process or non-CMP process, such as reversing current to remove copper, a closed loop feed back system in real time used to control the copper rate
removal or end point detection is still a must in the control process. Typically, CMP processes measure the thickness profile of copper. This profile is the key in applying pressure on the pad to achieve the removal
rate. Different profile thicknesses apply different pressures to remove copper at various rates. This results in an even surface or a final uniform thickness. See Figure 1 |
|
|
|
 |
|
Figure 1This thickness is around 1,500 to 2,000 Angstroms, while the thickness of copper after ECP is around 10,000 Angstroms.
The cross diameter thickness copper profile is generated using current metrology tools, such as 4-point probe, laser/sound, X-ray, or eddy current technology. This thickness profile data is provided to the CMP tool to
control the removal rate. At first, the removal rate is very fast up until the thickness approaches 1,500 Angstroms. It then slows down by applying less pressure and slurry, or slowing the spindle speed and applying
less slurry and pressure, or both. An optical technique has been used in endpoint detection in the CMP process. This technique uses light intensity to determine the thickness of copper and end point detection. With
the optical technique, there are, of course, both advantages and disadvantages. The advantages such things as low cost, small sensor size, and multiple sensors mounted on the pad to cover a large inspection area.
However, this technique also has some major drawbacks. First, based on the light intensity, the optical technique needs a transparent or semi-transparent layer to effectively measure the thickness. Most materials having
a thickness between 400 to 500 Angstroms are considered to be transparent. Second, the optical technique needs a clear lens to receive and transmit light. In the CMP process, particularly for copper, a very highly dense
material has a thickness of around 10,000 Angstroms, which is far from being transparent. Being dependant upon a transparent or semi-transparent layer, the optical technique cannot measure the thickness of the first
9,500 or 10,000 Angstroms of copper. To solve this problem, vendor tools use guesswork. First, they measure the thickness of the copper and then process the CMP at the recorded speed, pressure and time. Next, they stop
the CMP process and then measure the thickness of the remaining copper. As a result, the rate of copper removal can be established in their CMP tool and hopefully every single copper wafer will behave in a similar manor
as the one tested. During this copper removal state, the optical technique does not provide any information of the current removal rate of copper. Keeping in mind that while copper is removed, the thickness profile of
the entire surface of the wafer cannot be generated. This valued information, if available, can be fed to the motion control servo system to control pressure differences at different thickness, different spindle speed
and different slurry to achieve an even surface or uniform layer of copper at the final thickness of 1,500 to 2,000 Angstroms. If this final copper thickness still has topography, the pad will follow this topography. As
a result, copper in a high spot will remain on the wafer as a residual spot even after the rest of the wafer has been completely removed of copper. See Figure 2 and 3.
RESIDUAL COPPER SPOTS POST CMP 300 MM DETECTED BY TFEM |
|
Figure 3 Eddy current can fulfill all the needs of controlling the CMP
process if it can provide the information in real time. Some CMP vendors have used eddy current and succeeded in some degree, but still lack several critical points to completely control the CMP system, such as actual
real time measurement, multiple source of sensors, stable reading, accurate thickness of real time measurements, and large sensor size. The large size of eddy current sensors require more exclusive distance from the
wafer edge. The farther the distance from the wafer edge, the less sufficient copper information the eddy current sensor can provide.Spindle speed is another factor in the CMP process. The CMP tool needs a high
sampling rate to measure copper thickness during rotations as high as 800 to 1,000 rpm. For example, at a rotation speed of 800 rpm, it rotates 13.33 circles in one second or an equivalence of 12,479.5 mm (based on a
298 mm scanning diameter). If the sampling measurement is in the order of one sample per 100 milliseconds, by the time the thickness acquires the second sample, the actual measurement location has traveled away to
another distance of 124.79 mm. Andrew NDT Engineering, Corp. based in Gilroy, California has developed a patented non-contact measuring process that involves high speed eddy current sensing to measure the
thickness copper during CMP process in real time quickly and accurately. The company's Thin Film Electro Magnetic (TFEM) system is faster than optical and other existing eddy current systems integrated in the CMP tool.
The maximum sampling rate from TFEM is 100,000 samplings per second. It can measure the sample spot at up to 1.24 mm away from the first sample. Because the sensor size is 1 mm, the effective measurement area is almost
identical to the first measurement. With the advantage of size and a patented designed sensor, the TFEM sensor can measure up to 0.5 mm exclusive from the copper edge. Multiple sensors can be embedded onto the CMP
system. The working distance from the TFEM system can be a few hundred microns to 50,000 microns. Its unique designs can be mounted anywhere on the wafer of the CMP tool: above the copper side or under the backside of
wafer. A unique distance compensation due to pad erosion or different wafer thickness will maintain TFEM accuracy and repeatability. With thickness measurements in real time from any location on the wafer, a closed
loop feed back control system can be achieved to adjust pressure, spindle speed, and slurry. Andrew NDT Engineering's design team can modify sensors, hardware, and source code to suit all requirements either of current
or CMP tools under development to provide a real time closed loop feed back system. It will make the CMP tool completely controllable. |
|
