I want to do some proof-of-concept work in IC analogue IC design. I know that plenty of free circuit simulators exist (LTSpce, NgSpice etc.), however they are meant for PCB simulation and most perform poorly when attempting any serious work with transistor level designs. I know that this is mainly because they are supplied with elemental (MOS) device models that fail to capture the intricacies found in true silicon circuits.

My question is whether there exist some open source Process Design Kits (PDK), libraries of SPICE models, that one can use with any of the free simulators (or PSPICE, cause the OrCad-light version supplied by Cadence would suffice). I did some work in Cadence IC Design environments in uni and worked with both the AMS 0.35um pack, which is a real technology, and the GDPK 180 which they provide as a "general" PDK to play around. The later is exactly what I'd be looking for. For this project:

  • I don't need latest deep submicron tech (none of the SiO/FinFET stuff), it can honestly be quite old tech (like the aforementioned 180nm)

  • It doesn't have to be super-robust for any ultra-high frequency applications, I'm not intending on RF applications

  • It must be able to handle low power, subthreshold region operation (which is mainly why I'm asking the question)

  • It would be nice if it converged well with 100 nodes. From my experience PSPICE produced some very odd results (random jitter...) in circuits that contained multiple complex components (this might be the simulator's fault, the models' fault or my fault, hard to tell)

  • I would like to be able to specify gate width and length as a minimum for PMOS and NMOS devices, instead of having a discrete model for every possible W/L configuration (preferably).

I know PDK's are provided by foundries and are process specific. I know this is often industrial data that is protected as IP. All I hope for, is that for academic purposes there are libraries of models that aren't rubbish and are "good enough" if one wants to do serious simulation of circuits in CMOS.

Thank you for any responses!

  • \$\begingroup\$ What makes you think LTspice and ngspice are "meant for PCB simulation" and "perform poorly"? That's a strong accusation. Can you support it? \$\endgroup\$ – Elliot Alderson Oct 29 '19 at 14:00
  • \$\begingroup\$ SPICE was in fact developed specifically for IC design. \$\endgroup\$ – Dave Tweed Oct 29 '19 at 14:19
  • \$\begingroup\$ I would have to believe that such information is readily available through MOSIS. \$\endgroup\$ – Dave Tweed Oct 29 '19 at 14:22
  • \$\begingroup\$ It's not the SPICE programs themselves I have in mind, it's the resources that come with them. For example, LT gladly provides users with it's IC chip models (opamps, voltage regulators, etc.), but these are empirical models derived from component performance, AFAIK. These programs don't simulate single transistors that comprise their products, but merely specify the relationships between voltages and current on various pins. The MOS models they provide are either very basic or empirical models of specific power MOSFETs. \$\endgroup\$ – Olaf Sikorski Oct 29 '19 at 16:56
  • \$\begingroup\$ @OlafSikorski If you expect SPICE models, in general, to be accurate representations of reality, you're in for a great disappointment. \$\endgroup\$ – a concerned citizen Oct 29 '19 at 17:08

I only know of two so far:

  • \$\begingroup\$ Hi, it's been a while, but those are the kind of resources I was looking for. Thank you, hopefully this answer will help sb in the future too :) \$\endgroup\$ – Olaf Sikorski Dec 10 '20 at 20:02

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.