Current starved ring oscillator sensitivity to temperature

[Junus2012]

Hello,

below is the simple current starved ring oscillator,

I have removed the transistor Ts and simply sourced the circuit with ideal current source. According to the working principle, the charging and discharging time will be controlled by constant current source, and since the current is constant, then frequency of oscillation should be stable and not sensitive to temperature. However, the circuit during simulation is showing me sensitivity, what is the reason

If I want less sensitive to temperature clock oscillator will the fully differential delay stage would serve better, or you suggest me to move to more accurate scheme like other types of relaxation oscillator?

Thank you

 

[FvM]

I presume you can answer the question yourself by closely looking at the detailed simulation results. How are voltage and current waveforms changing over temperature?

 

[dick_freebird]

Control the tempco of the current sources to compensate the other drifts (Cj, u0, VT, ....)

 

[Junus2012]

I presume you can answer the question yourself by closely looking at the detailed simulation results. How are voltage and current waveforms changing over temperature?

Before doing simulation, transistors P2, P3.... and N2, N3, ..... from the circuit are mirrors of P1 and N1 respectively, where the latter are sinked/sourced with ideal constant current source. This mean the output of the current mirror should not change by temperature, which is by the end should keep the frequency insensitive to temperature.

--- Updated Aug 19, 2021 ---

trol the tempco of the current sources to compensate the other drifts (Cj, u0, VT, ....)

The equation of the frequency oscillation of the current starved oscillator is fosc=Id/(N C Vdd) and has no u0, VT variables to control, that is why I presume if I have constant current source then fosc shouldnt change

 

[d123]

Hi,

Why not simulate P1, N1 and the constant current source on their own and see if the 'constant' current going in and coming out has minute or not so minute changes over temperature? Is Dick pointing out that the schematic is idealized but the transistors will have real non-ideal responses and need temperature compensation of some kind? I don't understand what he or you are saying much but think I have a pretty good idea from disappointing results from so-called 'temperature-stable' current sources and voltage references. The hoped-for straight line always turns out to be a bit curved. Like simulating transistor-based op amp design current sources and sinks require rather complicated inverse mirrors of the mirrors to compensate for temperature drift?

I'll get my coat now, as they say.

 

[dick_freebird]

Point is that mobility, VT and burden capacitances
all have a temperature profile and the best you can
do is create an opposing influence that makes the
prime care-about, as flat as possible (while ensuring
that you didn't ruin something else along the way).

 

[BigBoss]

Since there is no Temperature Compensating Circuit, what would you expect ??
The Current Mirrors are Temperature Dependent even they are driven by Ideal Current Source.
You should find a Less Temperature Dependency Solutions first for Mirrors.Otherwise your Circuit will always exhibit this dependency.

 

[dick_freebird]

I'd note that a flat current mirror may not make
a flat center-frequency; the "right answer" might
be proportional to depletion capacitance (in turn,
proportional to temp?) or whatever. The first step
might be to characterize the delivered-current
profile that gives you what you want, and now
you've got your testbench starting from
behavioral to detailed.

 

[Junus2012]

Dear friends,

Thank you very much for your contribution to my questions and for your kind help,
I have followed your suggestions,
I have simulated the current mirror outside the ring oscillator to study their current dependency on temperature as per your suggestions, however, the simulation shows me perfect matching with negligible change in the current, I used an ideal current source of 20 µA, and the max change due to T was less than 1 nA, hence it can not be contribute to the change in the output frequency.
I conclude that the frequency shift in my oscillator was due to the temperature dependency of the inverters capacitors that used to create the delay.