jbdavid
|
Eugenes problem doesn't apply GENERALLY to Verilog-A models in Cadence's AMS Designer, its only a problem for Verilog-A Models called from the the spectre model files. and it just takes a little restructuring to pull the resistors/ diodes capacitors OUT of a verilog-a structural model into a subcircuit that calls that verilog-A model.. and this is one area where only the Verilog-A subset can be used.. Ie the model files have to work with spectre as well as with AMS designer..( and this might already be fixed - last time I had to fix this type of situation was about a year ago..)
When you get to the CIRCUIT LEVEL behavioral modeling, (veriloga views for circuit where you have a schematic ) The difference is that veriloga views (which can have structural as well as behavioral elements will work in Spectre (analog simulations) as well as Verilog-AMS (AMS DESIGNER) simulations..
AMS designer can handle both Verilog-A subset and Verilog (ie GATE, BEHAVIORAL or RTL) models.. But spectreVerilog could do that also.. What Verilog AMS gives you is the abilty to write MIXED SIGNAL models .. so that the ADC model has analog input connections from the "spice" side of the circuit, but LOGICAL output pins for the converted value, and a logical input pin for the clock.. if this Clock is one of the high speed signals in your circuit, it will dominate the spectre simualation.. but since its Logic in the AMS simulation, the model of the ADC is event driven, and actually uses an interpolated value of the input signal allowing only the filter in front to determine the analog solver's timestep. In a Functional verification environment this gets you the SAME functional behavior (voltage converted to bits) at potentially 50-1000x shorter simulation times... which means you can actually check ALL the control signals.. (Powerdown, enable, select Ain, selectBin ....and maybe use a verilog model of the dsp as well..
In a PLL circuit I worked with recently I was able to watch the PLL lock in a few minutes compared to hours watching it get close to lock with VerilogA behavioral models.. just by keeping the output of the VCO, and clock divider and input to the Phase detector in the Logic domain.. It does take the equivalent of 3 spectre licenses though..
The other advantage is that now I ALSO have all the power of ncsim.. including VHDL, VHDL_AMS support for IP, transaction based verification approaches VPI, and PLI integration, SystemC , PSL assertions (but only for the digital side right now) .
but a little partitioning of your design can minimize the need for real mixed signal simulations.. If all the control bits are coming from a register bank and are essentially static, you can write a simple verilogA model to drive the whole set, and you might not even need to simulate any control bit transitions.. in fact to support this case, I've got perl scripts to generate the stimulus block model, the verilog stub model (gets me a symbol) and the testbench and then import (verilogin) the testbench and stimulus block.. It worked fine on one recent tapeout of mine where the digital to analog interface was very clean.. I only had 1 set of 5 control bits that I had to worry about transitions on, and verilog-A was sufficient for that as well.. (testbench) .. but we have a dedicated systems team doing all the "full chip" behavioral modeling in matlab too.. (not linked to any actual schematics though)
AMS designer does have a way to link matlab models in also.. Hope this helps.
|