SPICE simulation -- how to bias a crystal oscillatoir

Numerical discretizing schemes generally don't preserve exact eigenfrequenc ies anyway. For instance, in FDTD electromagnetic simulations, you have to dork the value of c to improve accuracy--the numerical c can differ from th e real one by as much as a percent or so.

In an adaptive solver like SPICE, I'd expect the eigenfrequencies to depend on all sorts of internal variables such as time step, solver algorithm, an d a bunch of control panel parameters.

Cheers

Phil Hobbs

In LT Spice, the default sim of a simple ringing LC runs about 0.8% slow.

The crystal model, as well as the CMOS NMOS/{MPS models, along with the SPICE netlist is below -- full text format. .INCLUDE modelcardold.nmos .INCLUDE modelcardold.pmos

.SUBCKT INV 1 2 3

• 1 VDD
• 2 IN
• 3 OUT M0 3 2 1 1 M_PMOS L=1.5u W=2.0u M1 3 2 0 0 M_NMOS L=1.5u W=7.5u .ENDS

.SUBCKT CTAL 1 2

• 1 IN
• 2 OUT C0 1 2 20.0E-12 C1 1 3 5.0E-15 L0 3 4 0.5mH R0 4 2 1.0 .ENDS

VDD 1 0 DC 5.0 AC 0.0

C0 2 0 5p C1 4 0 100p C2 2 4 50p R0 2 3 10.0Meg R1 3 4 100 RL 5 0 50.0 XINV0 1 2 3 INV XINV1 1 3 5 INV XTAL 2 4 CTAL

.OPTIONS NOPAGE METHOD=GEAR .IC V(2)=0.3 V(3)=0.5 V(4)=1.0 V(5)=0.25 .PROBE .TRAN 100.0ns 500.0us 50.0ns UIC .PRINT TRAN V(5) .END

.MODEL M_PMOS PMOS LEVEL=3

.MODEL M_NMOS NMOS LEVEL=3

Yes, I am fully aware of the "kick start" method -- I have used it in the past for simulating free running multivibrators, ring oscillators etc., In fact, the netlist above uses the same trick in the SPICE .IC directive. Hopefully, the netlist would be helpful,

Am 07.11.2015 um 23:48 schrieb Phil Hobbs:

Xtal oscillators are easier analyzed in the ac domain and not with integration over time. Cut the loop open and calculate the loop gain. It will oscillate where the phase of the loop gain crosses 0. You will also see if it will compress ( amplitude of loop gain @ phase 0 >> 1) or not oscillate at all (amplitude there < 1.0) Source and load impedance at the cut point may be very different, and also different from 50 Ohms if you use s-params, so Randall-Hock correction may be needed to make up for the missing self-termination.

BTW you get weird results with LTspice for steps < 5 Hz or so in a

regards, Gerhard

The book they mention is good.

Cutting the loop open is not necessary. See...

Try the "Alternate" solver... the "Normal" solver takes liberties.

...Jim Thompson

That seems bass-ackwards... PMOS is usually 3-4X the width of NMOS.

Your small devices won't support a 50 Ohm load.

What is your .IC = {Value ?} ...Jim Thompson