Same differential testbenches give different results

[Frank]

I have a Spectre testbench to test my circuit which is a differential
transmit/receive switch. The testbench works well at the beginning.

After that, I copied that testbench to a new cellview and run the
simulation with this new cellview. Surprisingly, the new results are
completely different from the old one.

Is it because Spectre has problem dealing with differential circuit?
Please help me. Thanks.

 

[Andrew Beckett]

On 19 Jan 2006 00:18:01 -0800, "Frank" <chientm@gmail.com> wrote:

I have a Spectre testbench to test my circuit which is a differential
transmit/receive switch. The testbench works well at the beginning.

After that, I copied that testbench to a new cellview and run the
simulation with this new cellview. Surprisingly, the new results are
completely different from the old one.

Is it because Spectre has problem dealing with differential circuit?
Please help me. Thanks.

Well, something is different. Take a look at the netlists - spectre shouldn't
have any issues with this. Compare the input.scs from the first simulation to
the second, and hopefully you'll spot what's different.

Regards,

Andrew.

 

[Frank]

I checked the netlist and haven't found the difference.

In fact, I have two testbenches of one fully differential circuit on
one schematic. The operating conditions of these test benches are the
same. And the circuit itself is also symetrical between input and
output (i.e. the input and output are exhangable). Could this be the
source of unstable results? I mean whether the matrix solver have
trouble with a matrix that has a lot of symmetry?

Frank

 

[Andrew Beckett]

On 20 Jan 2006 01:22:26 -0800, "Frank" <chientm@gmail.com> wrote:

I checked the netlist and haven't found the difference.

In fact, I have two testbenches of one fully differential circuit on
one schematic. The operating conditions of these test benches are the
same. And the circuit itself is also symetrical between input and
output (i.e. the input and output are exhangable). Could this be the
source of unstable results? I mean whether the matrix solver have
trouble with a matrix that has a lot of symmetry?

Frank

The matrix solver wouldn't care about symmetry in the design. It's possible of
course that you have multiple stable operating points, which could mean that
you're sensitive to starting conditions - even the order in the netlist could
affect that.

Andrew.

 

[Frank]

I think I figured out the problem.

Because I am using the NMOS as an RF switch, I biased the source and
drain terminals of the NMOS to the same DC potential by RF chokes. This
makes VDS=0. It seems that the simulator does not happy with VDS=0 so
that the simulated RF performance fluctuates between some undefined
states. I made VDS=1f V and the fluctuation disappeared.

At least, at the meantime, the simulation results are reasonable. I am
not sure that whether VDS=0 could cause any problem to the BSIM3 model
or Spectre or not. My NMOS is operating as a switch, its DC operating
conditions should be easily settled down by the simulator (except when
VDS=0, I guess)

Frank