Let me try to clear a few things up. First, this is not a convergence problem, and it is not a Spectre problem. Your circuit is unstable and Spectre is right to issue the blow up message because the resulting waveforms do shoot off to infinity.
Your circuit is unstable because your inductance matrix is not positive definite. I know this because I converted your netlist into a python script that constructed and analyzed the eigenvalues of your inductance matrix. That script is:
Code:import numpy
from math import sqrt
L = numpy.zeros((10, 10))
L0=1.3e-9
L1=1.3e-9
L2=1.4e-9
L3=1.4e-9
L4=1.5e-9
L5=1.5e-9
L6=1.265e-9
L7=1.265e-9
L8=4.165e-9
L9=4.165e-9
L[0][2] = 1*sqrt(L0*L2)
L[2][0] = 1*sqrt(L2*L0)
L[1][3] = 1*sqrt(L1*L3)
L[3][1] = 1*sqrt(L3*L1)
L[5][7] = 1*sqrt(L5*L7)
L[7][5] = 1*sqrt(L7*L5)
L[4][6] = 1*sqrt(L4*L6)
L[6][4] = 1*sqrt(L6*L4)
L[7][8] = 0.185*sqrt(L7*L8)
L[8][7] = 0.185*sqrt(L8*L7)
L[6][9] = 0.185*sqrt(L6*L9)
L[9][6] = 0.185*sqrt(L9*L6)
L[0][7] = 0.823*sqrt(L0*L7)
L[7][0] = 0.823*sqrt(L7*L0)
L[1][6] = 0.823*sqrt(L1*L6)
L[6][1] = 0.823*sqrt(L6*L1)
L[0][0] = L0
L[1][1] = L1
L[2][2] = L2
L[3][3] = L3
L[4][4] = L4
L[5][5] = L5
L[6][6] = L6
L[7][7] = L7
L[8][8] = L8
L[9][9] = L9
ev = numpy.linalg.eigvalsh(L)
print ev
It computed the eigenvalues as
Code:[ -6.69232309e-10 -6.69232309e-10 4.50778482e-10 4.50778482e-10
2.31985763e-09 2.31985763e-09 3.26523077e-09 3.26523077e-09
4.26336543e-09 4.26336543e-09]
Several of the eigenvalues are negative, meaning that the inductance matrix is not positive definite, meaning that your coupled inductor is not passive.
So this leaves three questions to be answered.
1. Why does ADS eigenanalysis report the circuit as being stable?
I don't know why. Perhaps Pancho Hideboo accidentally modified the circuit, or perhaps there is a bug in ADS. A pole-zero analysis by Spectre reports RHP poles.
2. Why does harmonic balance analysis give a bounded result (even with k=1)?
Like DC analysis, HB analysis computes a steady-state solution regardless of whether that solution is stable. In a very real sense, this is a problem with HB analysis because it can lead you to believe your circuit is working properly when it is really unstable.
3. Why do HSpice results remain bounded?
I believe that this is because HSpice is using too large a timestep. It is a known problem with transient simulators that like HB simulators they too can sometimes ignore instabilities if they take too large a timestep. Forcing them to take smaller timesteps will eventually cause them to properly produce waveforms that exhibit the instability.
One final comment. In his last post, Pancho Hideboo seems to suggest that the fact that HB simulators are capable of simulating non-physical systems is a good thing. In fact it is not. It can lead designers to into a false sense of security. For example, in this case if you were using only a harmonic balance simulator, you might have missed the fact that your inductor model is fundamentally flawed.
-Ken