Method for Accurate Plating Control In Pulsed Power Supply Applications
Original Publication Date: 2000-Dec-01
Included in the Prior Art Database: 2003-Jun-19
Many commercial plating power supplies are set up for asynchronous sampling. When running in pulse plating mode, this asynchronous method is not accurately reporting the true current that is being applied to the wafer, substrate, or other plated object. This is especially true when using long pulse widths for alloy plating. Even supplies with control sampling algorithms that are synchronized to the leading edge of the pulse do not always handle long pulse widths well. Since the power supply controller knows the value of the pulse width that is being used for each plating operation, it would be a matter of calculation by the controller to set the sampling point to the mid-point of the pulse width. This mid-point would be more of a true measure of the total coulombs being applied to the plated object, than that which can be obtained from any other calculated method. With longer pulse widths, this is even more critical. Many of the existing power supply designs have a sampling algorithm which is asynchronous, which can be seen in the current fluctuations being reported. If the averaging function in the algorithm is sufficiently long, this fluctuation cannot be seen in the reporting function, but can be seen by comparing an external coulomb meter with what is being reported. Even synchronous supplies can have sampling problems with long pulse widths, as most use a fixed sampling time triggered from the leading edge of the pulse. With only one sample point during long pulse widths, inaccurate results can be obtained if there is any signal disturbance after the sample is taken. Using an external coulomb meter in series in the circuit shows that the current being supplied by the power supply is constant and steady, but the reported value can fluctuate as the sample is taken during the rise and fall times, and even during the "off" period of the cycle. These values can be seen in low, intermittent, and zero values being reported. The result of these incorrect reported coulomb values is a plated layer thicker than what is desired. If one is trying to control the plated thickness by coulombs delivered to the wafer, then corrections would have to be guessed at as each plating run would not behave as expected for the thickness vs. coulombs delivered. Additionally, since the fluctuations are not fixed or repeatable, the results would vary somewhat with each plating run, further compounding the problem of control. The alternative is to plate by time, which does not give ideal thickness control compared to coulomb control. By utilizing the data available to the controller and sampling at the midpoint of any width pulse, as triggered from the leading edge of the pulse, a more accurate coulomb control of plated thickness is achieved. Even in synchronous supplies, long pulse widths can have sampling points taken several times over the length of the pulse when the controller is fully utilized. This method would result in the most accurate control scheme, as verified with external coulomb meters.