In this interview, members of the Texas Instrument wireless group talk about the characterization effort initiated and completed last year between Cadence and IBM using TI RF designs as a pilot. The goal between the two teams was to optimize SpectreRF usage to successfully and efficiently simulate TI RF oscillator designs. This tight collaboration resulted in a publication at DesignCon conference (1). This characterization was successfully implemented in MMSIM 7.2 and above.
The people involved in this work are:
Q: Nand, Ted, can you please give us some background about RF design simulation and the flows and procedures you are using?
Ted (photo, right): In the design of IC transceivers and SOCs for wireless applications, we use Spectre for a variety of simulation tasks, including transient, transient noise, small-signal and operating point analysis. For front-end design we use SpecteRF HB (Harmonic Balance) for IP and NF analysis, as well as oscillator PSS and noise analysis. In the development process, all simulations are performed on schematic views at the module and top level, then re-run using Assura QRC extracted netlists after layout.
Running process-corner, temperature, and voltage sweeps and even statistical analysis on large RC extracted circuits requires a tool that is as fast as it is accurate, as simulation time is often a limiting factor in terms of what can be achieved under the tight schedule constraints that we commonly face. My team and I depend heavily on the accuracy, flexibility, and simulation throughput of Spectre and SpectreRF in almost every analog/RF simulation task performed at TI.
Nand (photo, left): We use Spectre for various analog and RF circuit simulations. We use it for transistor level simulations and perform various required analysis such as DC, OP, transient, and AC analysis. In high frequency domain pss (HB or Shooting), Pnoise analysis also comes into picture. For supply rejection we currently use PXF.
Q: What simulation challenges are TI design engineers facing with oscillators, especially when dealing with advanced process nodes and complex architectures?
Ted: Many tools can simulate the inherent phase noise of an LC oscillator core with reasonable accuracy, but often the results become less reliable when external sources such as bias and supply noise are considered. Also, simulating high-Q crystal oscillator circuits is often a particular challenge. In past designs, it has taken a huge amount of simulation time to arrive at a reliable PSS solution before PNoise results become completely trustworthy.In circuits with amplitude control loops, integrated reference and supply regulators, buffer and divider chains that follow the oscillator and strongly affect noise and signal swing, even a relatively simple simulation can be very challenging when all of these factors are considered.
Nand: We used to have convergence issues with high inductors. Also, for supply rejection we are struggling to use multi tone HB analysis with autonomous circuit. The circuit converges easily in single tone mode but we face challenges while trying to simulate multi tone.
Q: What was the triggering factor that started this collaboration?
Ted: For some time now, we have depended on SpectreRF's Shooting Newton algorithms for oscillator analysis, but HB engines clearly have some advantages in terms of simulation time and accuracy. When Cadence made HB a primary part of their suite, we jumped at the chance to try it out, but we also had to build confidence in the tool in a variety of scenarios before relying on it exclusively.
TI's ISM band transceivers and SOCs in particular have extremely stringent close-in phase noise requirements, which means that every possible non-ideal element has to be carefully considered, accurately modeled and simulated. The collaboration was a huge factor in bringing us up to speed with the updated features of the tool, ensuring that all results matched with theory and expectation, and reducing simulation time without compromising accuracy.
Nand: Due to stringent close-in phase noise requirements in our circuit, we had to consider every possible non-ideal element and it has to be accurately modeled and simulated. However HB phase noise had some higher dependency on tstab and few simulation settings. The collaboration helps to match results with theory and expectation, and reducing simulation time without compromising accuracy. This also helped to remove dependency on tstab by considering oversample into simulation algorithm.
Q: Can you describe the different steps that TI and Cadence took jointly to optimize SpectreRF usage for TI wireless designs?
Ted: The collaboration began as an investigation of HB vs. Shooting Newton for LC VCO noise analysis, and later expanded to include XOSC noise analysis and even multi-tone simulations for PSS supply-rejection in our low-jitter XOSC core and buffer chain. Initially, we noticed some unexpected dependency on certain parameters such as tstab in terms of the noise results that were reported, especially close-in where reference and regulator noise became significant contributors.
TI and Cadence worked closely, passing test circuits and results back and forth to determine the best way to optimize parameters for improved throughput and consistency of the results. This resulted in an enhancement of phase-noise algorithm for both shooting-Newton and HB. The enhancement was delivered in MMSIM7.2 and gave believable and consistent results. In addition, we had a strong desire to separate AM from PM noise, which in Shooting was done by choosing method=modulated, but this was not as straightforward with HB. The Cadence R&D team also implemented a fix for this limitation, and expanded the conditions under which this option could be used.
Over the course of a few months, TI design and Cadence Support and R&D teams met weekly to track progress on various issues, and also shared a great deal of ad-hoc communication as specific questions or concerns arose. In the end, we were able to simulate the LC VCOs under development, and also compare a previous-generation transceiver's measured results with the results from the updated version of SpectreRF and achieve good correlation.
Finally, the Cadence team ran transient noise analysis on our complete test circuit in Spectre and showed that these results also correlated very well with HB Noise, which is something that I have not seen before from other vendors.
Q: Can you describe the outcome and results? What was achieved through this collaboration?
Ted: The result of the collaboration was to greatly improve our understanding of and confidence in Spectre's HB tool when used for oscillator analysis. Again, the ISM band transceiver that was the test vehicle for most of this work performed in silicon as expected, which speaks volumes about the utility and accuracy of the tool.
We achieved very good simulation throughput in terms of corner and statistical analysis across a wide variety of supply, temperature, power/performance modes and of this VCO which combines analog and discrete tuning to cover a wide range of output frequencies for multi-standard operation. In addition, the Cadence and TI teams collaborated on multiple technical presentations, both at CDNLive09 and DesignCon 2010.
Nand: Simulation speedup, higher accuracy, and a stable HB flow. This helped us to reduce design cycle time and increase yield.
Q: How is this Cadence-TI collaboration going to improve TI productivity and success?
Ted: The improvements realized in Spectre HB and our understanding of the tool has made it a completely viable option for tackling many of our most difficult simulation tasks. It is also a very cost-effective solution compared to competing RF simulation tools.
Q: What was your overall experience working with Cadence team?
Ted: The Cadence team was extremely responsive and worked tirelessly to ensure that we understood the most efficient and accurate ways to use Spectre HB and were satisfied with the tool. Their technical competence and the energy they devoted to ensuring our success were huge assets.
Nand: The Cadence support team was very helpful in providing us quick solutions for issues. They understood issues quickly and gave us a stable solution. We were extremely satisfied.
Q: How do you see this collaboration in the future? What would be the next step?
Ted: I believe there are further tweaks that can be implemented for PSS supply-rejection simulations, both in terms of ease of use and perhaps even accuracy, that we may pursue in the future. Our close ties with the Cadence Support and R&D teams should make any additional steps quite easy to ramp.
Nand: We expect Cadence will help us to solve future problems with the same enthusiasm.
Publications:(1) "Phase-Noise Characterization for RF Oscillators with Bandgap Reference and Isolation Buffer", Yu Zhu, Helene Thibieroz, Nand Jha, Ted Blank, DesignCon, 2010