Spice simulation, Tina-TI, LTSpice, PSpice, and more
I got a nice note from the irrepressible Mary Dunnie that Texas Instruments has improved its Tina-TI Spice tool. It is available in simplified and traditional Chinese, Japanese and Russian, as well as English. From the release:
- Support for multi-core processors and an optimized SPICE engine enable TINA-TI 9.1 to run simulations 5X faster on average.
- Designers can import any SPICE model to easily simulate their designs in TINA-TI 9.1.
- TINA-TI 9.1 will feature more than 500 part models and reference designs including more than 130 new power models. The broad model coverage and fast simulation in TINA-TI 9.1 make it easier for designers to speed their products to market.
- TINA-TI 9.1 provides all the conventional DC, transient and frequency domain analysis of SPICE and extensive post-processing capability that allows the user to customize results. Virtual instruments allow users to select input waveforms and probe circuit node voltages and waveforms.
My buddy Martin Cano over at Agilent is buying the paid-for version of Tina-Spice and I will let you know how he likes it. He was the one that helped me simulate op-amp noise with signal generators. Trick is the signal is complex, so you have to take the nV/rtHz results and take the absolute value first before you integrate it to get uV of RMS noise over a bandwidth. I will always respect Cano for figuring that out. I know the RMS noise could not be going down over bandwidth like PSpice was showing, but darned if I could have figured it out. Taking the integral of the ABS of the noise spectrum did the trick. And no, this does not help simulate 1/f noise, that is a whole ‘nuther problem.
I have learned a few things about Spice lately. A buddy that retired from Linear Technology explained why Mike Engelhardt was so brilliant when he developed LT’s free LTSpice. See, LT is famous for their power converters. Power converters are notoriously hard to simulate. I thought it was because of the problems simulating magnetics, but there is another fundamental problem. It may take milliseconds or even seconds for the circuit to come up to its final steady-state operation. If your Spice engine is doing its matrix math every 20 nanoseconds, than means it will take a long long time to solve the circuit. Phase-locked loops have the exact same problem. You can use harmonic balance and other RF steady-state frequency domain tools to look at the steady-state operation, but it is critical to see how the PLL starts up and pulls into frequency lock. The same goes for switching power converters.
So what Mike Engelhardt did was figure out a way to radically speed up the solution of the circuit as it comes up to its steady-state value. He personally writes the models for the switching regulator converter chips so that they work with his algorithms. Things I thought were hard about power converter Spice, like modeling the magnetics, are pretty well understood, as long as you good Coilcraft models or similar. So now many of the high-dollar Spice packages have fast solvers to help in PLL design, but they don’t do anything about power converter IC models so you can’t use them for that. Linear Tech and Mike Englehard solved an intractable Spice problem over a decade ago that the rest of the EDA community is still catching up to.
This also explained why I have been confused about the openness of LTSpice. People would tell me that it only runs with LT parts. I figured it was a closed system that only runs LT parts. Well, yes and no. Since Mike did not write the models for TI parts, they cannot take advantage of all his clever start-up algorithms. I assume a TI power part will run in LTSpice, but only like it would run like in any other Spice package that does not have Mike’s cool algorithms. (Mike based his work on PSpice, but most of the code is his).
But here was a big a deal I just realized about LTSpice. It will run a model from any company’s parts. When it comes to op amps, LTSpice will work just fine with any model. And just like commercial high-dollar Spice, if you put in crappy old models like the LM393 you will get crappy results. If you use those CLC models National Semi got from Comlinear, well Mike Steffes (now at Intersil) made sure those op-amp models were almost like transistor-level macro-models.
OK, Anther great thing I learned recently about Spice came when I met Dr. Richard K. Oswald at the embedded systems conference. He is a Spice expert and has been involved since it was Berkeley SPICE. He started working in disk-drive read channels and lately spends his time with switching regulators and patent work. He is a heavy-hitter and he just taught me that PSpice uses proprietary models that don’t run on the open-source Spice packages derived from Berkeley SPICE. And everything, every single Spice on earth, does come from Berkeley’s work. Some companies just hang a schematic capture program in front of the Berkeley engine, to feed it a netlist. I met a PSpice guy once and he said all their effort was in making things converge. Some of my pals still like the old PSpice schematic editor better than Orcad, go figure.
And while on the subject of Spice-run convergence, a big shout out to Matt Berggren over at Altium. He was the guy that explained that we have to remember that SPICE stands for “simulation program with IC emphasis”. So when us board-level folks use it, we can’t let the defaults run, or it may not converge. Matt recommended an out-of-print book, Ron M. Kielkowski’s Inside Spice, also available a little cheaper as an earlier edition, which Matt says has all the same important info.
Finally a good article from Don LaFontaine over at Intersil about models for low-noise low-power op amps. Also a blog I did about Tamara Schmitz and Jian Wong referring to an article they wrote in EETimes, back when we were competitors. Now Bill Schweber and I both work for Ron Wilson so you get double the analog goodness. Here is an article I did about on-line Spice tools, which did not talk a lot about LTSpice since it is a downloadable program, not an on-line one. I would be curious to hear from you if you prefer on-line tools like WEBENCH or downloadable tools you install like LTSpice.