SystemC based Behavioral Modeling and Virtual Prototyping of RF SoCs

Aachen (2017) [Dissertation / PhD Thesis]

Page(s): 1 Online-Ressource (xx, 136 Seiten) : Diagramme

Abstract

Today’s multi-standard multi-band mobile terminals are enabled by the increased complexity of RF-SoCs. Multi-band multi-mode RF frontends based on DigitalRF-architectures using extensive self-calibration/compensation schemes have to be realized in nanoscale CMOS nodes. Pre-tapeout verification is a key challenge for this kind of RF-SoCs as conventional circuit/transistor-level simulations using Fast-Spice are not able to cover the requirements. The only left practical solution is the behavioral modeling based method based on event driven data flow simulation techniques. The purpose of this thesis is to enhance the effectiveness of the behavioral modeling method in functional verification of RF SoCs by introducing a RF virtual prototyping (RF VP) technique. The idea of RF VP is to use pin-compatible behavioral models to replace RF and mixed-signal parts of the RF SoCs. The models are required to be digital simulator supported, and close linked with the latest snapshot of the circuit behaviors. This technique aims to provide high simulation efficient and accuracy-guaranteed solution to RF SoCs verification. It involves how to describe analog behaviors in digital way, how to increase simulation speed by switching to another signal abstraction, how to reduce manual modeling efforts and human errors by self-developed EDA tools, and etc. In this thesis, SystemC is chosen as modeling language for the RF VP due to the high simulation efficiency, the syntax flexibility, and the good integration into digital RTL simulation tools. Most of all, it is selected due to its order of freedom, which is referred to the ability to cross levels of abstractions of the models. Higher order of freedom means more space to make the trade-off between simulation speed and verification accuracy/coverage, which is the basic trick in response to the huge increase of the complexity. Further discussions are expanded on the detailed behavioral modeling methods using SystemC. With event-driven method, the analog behavior can be modeled as a signal flow without solving differential equations. In order to speed up simulations, an improved baseband modeling method covering up to 3rd order harmonics is used. The model can be switched to baseband signal abstraction without carrier distortion. This baseband method is further extended to carry a vector of spectral components: more accurate but still high simulation efficient. By using a mixed-domain modeling method, a new signal abstraction is invented to bridge the behavioral description in baseband time domain and baseband frequency domain. To certain structure like PLL with feedback path, an event-shooting method is applied to compensate the unexpected feedback path delay when using the baseband signal abstraction. The case in modeling supply noise reveals that these modeling methods in SystemC can extend the modeling coverage from the main circuit functional behaviors to some other performance related behavior. SystemC RF VP technique is enabled on the basis of these modeling methods. Further, a set of self-developed toolsets/concepts are introduced to automate the RF virtual prototyping process. A lot of handcrafting efforts are saved and moreover many human errors are avoided by these toolsets starting from model netlisting, block level modeling/self-checking, model parameters extraction, to RF VP models generation and system simulation management. Two industry level test cases are used as the applications of the proposed methods in functional verification of RF SoCs. One is for a GNSS receiver front-end. Another is for a short-range low energy transceiver front-end. From the simulation results, these techniques can achieve a great simulation speed up (maximum to 105 speed up comparing to SPICE based analog simulations), while the simulation correctness is still reserved and the modeling effort is greatly reduced.

Authors

Authors

Chen, Zhimiao

Advisors

Heinen, Stefan
Ascheid, Gerd

Identifier

  • REPORT NUMBER: RWTH-2017-09469