Spectre doesn't have any "presumptions". However, floating nodes present a problem for circuit simulators, because they are trying to solve a system of simultaneous nonlinear differential equations, and a floating node is uncontrolled - you have an infinite number of potential solutions. Put another way, the matrix being solved is ill-conditioned. So what happens is that spectre identifies various situations that present floating nodes - either across various junctions that could be off, but also to ensure there is a path to ground. If I create a simple netlist that leaves both the ps and ns nodes floating: // vdc (in1 0) vsource dc=1m GAIN (in1 0 op1 op2) Vcvs gain=10.0 ahdl_include "forum50.va" dc dc (forum50.va is your veriloga model), spectre then tells me: No DC path from node `op1' to ground, Gmin installed to provide path. By doing this, a solution can be found - if op1 is 0V, no current flows through the resistor and hence no current flows through the voltage source to the other end of the vsource, and so op1 is 0V, and op2 is -10mV. The node that the gmin gets inserted at in this situation would be somewhat arbitrary - it just depends on the order the simulator processes the design (which you have no control over). Since if the output is floating, any absolute voltage for op1 would be feasible, it's a correct solution - just as correct as any others. If you'd tried to aid the simulator by (say) adding 1Tohm (that's the gmin resistance that normally gets added) to each of op1 and op2, then you'd get op1=5mV and op2=-5mV (this ensures that the current through both resistors is the same). So it's not a matter of "assumptions" or "presumptions", but spectre trying to ensure that a solution can actually be found. Regards, Andrew![]()
↧
