Neural modeling of dynamic systems with nonmeasurable state variables

2007 18th International Zurich Symposium on Electromagnetic Compatibility

Savi

et al. 2007

In this paper enhancements of parametric behavioral models for the output buffers of digital ICs are explored. A model based on a single-piece structure, which offers improved accuracy in describing state transition events for arbitrary load conditions, is proposed. This model exploits the potentiality of local-linear state-space parametric relations. These relations can be effectively estimated from input-output port responses only, and provide better stability properties and improved efficiency.

Section: Numerical Examplementioning

confidence: 65%

Single-piece state-space behavioral models for IC output buffers

2007 18th International Zurich Symposium on Electromagnetic Compatibility

Savi

et al. 2007

“…For an additional performance evaluation, we addressed ourselves to the assessment of model stability, by means of an analysis of the eigenvalues of the linearized model. The eigenvalues are computed for each point explored by the voltage and current responses during the transient simulations of a validation test [11]. Figures 2 compares the eigenloci of the linearized model equation for the dynamic component of iH, for ESN (circles) and NARX (crosses) 67 F models.…”

Section: Numerical Resultsmentioning

confidence: 99%

Guaranteed Locally-Stable Macromodels of Digital Devices via Echo State Networks

2006 IEEE Electrical Performane of Electronic Packaging

Maio

et al. 2006

The assessment of signal integrity effects in high-speed digital systems requires accurate and efficient IC macromodels. The proposed methodology is based on parametric relations that are expressed in terms of discrete-time Echo State Networks. This approach overcomes the stability limitations of traditional parametric macromodels used so far. Applications of Echo State Networks to the modeling of real devices exhibiting a complex dynamical behavior are discussed.

“…Finally, model stability is assessed by computing the eigenvalues of the linearized model (1). The eigenvalues are computed for each time step of the HSPICE simulations of the validation test, as suggested by [22]. Table I collects the main figures of the estimated models illustrated above.…”

Section: Performance Assessmentmentioning

confidence: 99%

Locally-Stable Macromodels of Integrated Digital Devices for Multimedia Applications

IEEE Trans. Adv. Packag.

Maio

et al. 2008

This paper addresses the development of accurate and efficient behavioral models of digital integrated circuits for the assessment of high-speed systems. Device models are based on suitable parametric expressions estimated from port transient responses and are effective at system level, where the quality of functional signals and the impact of supply noise need to be simulated. A potential limitation of some state-of-the-art modeling techniques resides in hidden instabilities manifesting themselves in the use of models, without being evident in the building phase of the same models. This contribution compares three recently-proposed model structures, and selects the local-linear state-space modeling technique as an optimal candidate for the signal integrity assessment of data links. In fact, this technique combines a simple verification of the local stability of models with a limited model size and an easy implementation in commercial simulation tools. An application of the proposed methodology to a real problem involving commercial devices and a data-link of a wireless device demonstrates the validity of this approach.