MIM cap is better nodoubt, put one can have poly-poly cap also.
The problem is area consumed by MIM cap. If I remember well Dr Shanthi Pavan from IIT Madras in his PhD work has talked on use of MOS capacitor in filter design. "HIGH FREQUENCY CONTINUOUS TIME FILTERS IN DIGITAL CMOS PROCESSES "
Following is abstract of it
"This dissertation investigates the design of high frequency continuous time filters in digital CMOS processes. A technique called ``constant-capacitance scaling'' has been proposed to make possible the realization of continuous time filters which are tunable over a very wide range, while dissipating low power and realizing very high pole frequencies. It is shown that filters designed using this technique are optimal with respect to noise, distortion and dynamic range. The architectures developed utilize MOS capacitors as the integrating elements. This enables the use of a fully digital CMOS process. By careful modeling, it has been shown that distortion generated in the inherently non-linear MOS capacitors is not of concern for a variety of applications. An accurate model developed for accumulation capacitors has been implemented in a general purpose circuit simulator.
In order to demonstrate the power of constant-capacitance scaling, a fourth order Butterworth filter has been designed and fabricated in a 0.25 micron digital CMOS process. This filter uses scaling techniques along with MOS accumulation capacitors and has a bandwidth which can be programmed from 60 MHz to 350 MHz. The measured dynamic range of the filter is 54 dB. In order to keep the response of the filter stable over process and temperature variations, a resistor-servo tuning technique which exploits the square law nature of the MOS transistor has been developed. This method is sufficient for filters with low quality factors.
For very high-quality filters, the simple tuning technique used in the test chip may not guarantee sufficient precision. A more accurate and complex method based on the behavior of a filter-comparator oscillator is proposed. This scheme converts a filter into an oscillator. Analytical relations for the behavior of the filter-comparator system have been developed and verified through a breadboard prototype. The amplitude and frequency of oscillation are measures of the center frequency and quality factor of a biquadratic section. The novelty of the filter tuning loop proposed is that it decouples center frequency and quality factor measurements."