In Reply to: Stable low freq on chip oscillator posted by Shripadaraj on 02/21/02 at 7:34 AM:
So, that's better than +/- 0.025% (250ppm) including initial tolerance, temperature effects, supply tolerance, and ageing, right???
If I knew how to make a monolithic oscillator that good I'd patent it, and hope to collect substantial royalties.
I'd think in terms of the fundamental quantities that can be synthesized in a monolithic circuit. To synthesize an absolute time interval, you'd need to have other absolute standards on-chip that are more accurate than 250ppm. One of these can be voltage, using a bandgap circuit. But to obtain time, you'd need to use this in conjuction with some other standard, such as capacitance; but I don't think an on-chip capacitor will be stable enough, even if it is possible to trim the initial value in a post-manufacture calibration process. (Actually, you'd probably trim the voltage reference to compensate for the initial capacitor value).
You might be able to take this a step further, and check how on-chip capacitors vary with temperature. If the temperature coefficient happens to be predictable, you could then devise a band-gap circuit having an equal but opposite tempco. But I don't know how enough about on-chip capacitors to judge how feasible this is.
Is the process you're using CMOS or bipolar, digital or analogue? Are you allowed to use an on-chip trimming technique (e.g. zener zapping)? The actual frequency is not an issue, since you can always use a divider (or PLL) to get the actual frequency you want. These questions all have a bearing on how feasible your objective is.
Linear Technology have a SOT-23 resistor programmed oscillator, the LT1799. Its accuracy over the commercial temperature range is +/- 2% plus the resistor tolerance.
Just my 0.02 worth, I'd be interested to see what other responses you get to your inquiry.