Hello all,
            I am designing a Low power crystal oscillator to operate at three different frequencies 10khz, 32.768khz and 100khz. I am using the circuit proposed in Eric Vittoz "CMOS Analog Integrated Circuits Based on Weak Inversion".  

1. What is the best way to start up this circuit.
2. This is a 1977 paper, I am not sure if there are any problems with this architecture or Do I need take care any issue specifically.

Thanks,
Ramakrishna

Hello Vivek,
 Thanks for the reply.
 
  The problem I have is that it is taking lot of simulation time for the amplitude of oscillations to get stabilized. I had given a current pulse on one of the nodes of the cystal model which starts the oscillations and the amplitude of the pulse applied is determining the initial amplitude of oscillation and thus simulation time that the oscillator's output takes to settle to its regulated value.    
  So I have to do lot of iterations to find out the right pulse amplitude that starts the simulations close to the final amplitude. This is time taking.  

My question is: Is there any better way to do this.

2. After the Voltage regulated output of the oscillator settles in simulation I am finding some kind of variation in the amplitude of the order of 100 uV, I think this is because of the finte impedance of the transistor biased in weak-inversion which alongwith a capacitor acts as a filter whose output goes to amplitude regulation circuit.  I think this is going to show up as jitter in the Square wave that is produced using the regulated oscillations.

My question is: Did you see this kind of effect during your simulations. What care did you take during design and layout of that transistor that gives us >100Mohms of resistance in weak inversion.

   Thanks for the information and experience you are sharing. It will be of great value to me.

Ramakrishna

Hi Ramakrishna,

Yawei is right. You really have to live with the slow simulation. A little improvement may be obtained by using a lossless integration method such as "traponly" if you use Spectre. If your AC calculation shows that you fulfil the startup criteria, then usually startup itself should not be an issue.

About the 100uV variation in amplitude, I am not so sure if I saw such a thing in my simulation. If it is indeed numerical error, then it is quite big. You should try to tighten the simulator accuracy options and see if it is reduced or if it goes away.

To my understanding, if the amplitude varies without causing any frequency variation in the crystal output, then you would not see jitter due to this if you convert it to a digital output. However, the amplitude variation indirectly changes the frequency of the crystal.

Vpk -> Gm -> Fosc.

The dependence is weak but it is there and you need to know how much jitter you can tolerate.

A well-designed oscillator should not really show much variation. You should try to stabilize the amplitude better. I believe the techniques are discussed in the papers I mentioned earlier. Vittoz mentions these and shows some curves for the amplitude control.

Good Luck
Vivek