This paper presents a comprehensive strategy for fast generation of passive macromodels of linear devices and interconnects on parallel computing hardware. Starting from a raw characterization of the structure in terms of frequency-domain tabulated scattering responses, we perform a rational curve fitting and a postprocessing passivity enforcement. Both algorithms are parallelized and cast in a form that is suitable for deployment on shared-memory multicore platforms. Particular emphasis is placed on the passivity characterization step, which is performed using two complementary strategies. The first uses an iterative restarted and deflated rational Arnoldi process to extract the imaginary Hamiltonian eigenvalues associated with the model. The second is based on an accuracy-controlled adaptive sampling. Various parallelization strategies are discussed for both schemes, with particular care on load balancing between different computing threads and memory occupation. The resulting parallel macromodeling flow is demonstrated on a number of medium- and large-scale structures, showing good scalability up to 16 computational cores.

High-Performance Passive Macromodeling Algorithms for Parallel Computing Platforms / Chinea, Alessandro; GRIVET TALOCIA, Stefano; Olivadese, SALVATORE BERNARDO; Gobbato, Luca. - In: IEEE TRANSACTIONS ON COMPONENTS, PACKAGING, AND MANUFACTURING TECHNOLOGY. - ISSN 2156-3950. - STAMPA. - 3:7(2013), pp. 1188-1203. [10.1109/TCPMT.2013.2257193]

High-Performance Passive Macromodeling Algorithms for Parallel Computing Platforms

CHINEA, ALESSANDRO;GRIVET TALOCIA, STEFANO;OLIVADESE, SALVATORE BERNARDO;GOBBATO, LUCA
2013

Abstract

This paper presents a comprehensive strategy for fast generation of passive macromodels of linear devices and interconnects on parallel computing hardware. Starting from a raw characterization of the structure in terms of frequency-domain tabulated scattering responses, we perform a rational curve fitting and a postprocessing passivity enforcement. Both algorithms are parallelized and cast in a form that is suitable for deployment on shared-memory multicore platforms. Particular emphasis is placed on the passivity characterization step, which is performed using two complementary strategies. The first uses an iterative restarted and deflated rational Arnoldi process to extract the imaginary Hamiltonian eigenvalues associated with the model. The second is based on an accuracy-controlled adaptive sampling. Various parallelization strategies are discussed for both schemes, with particular care on load balancing between different computing threads and memory occupation. The resulting parallel macromodeling flow is demonstrated on a number of medium- and large-scale structures, showing good scalability up to 16 computational cores.
File in questo prodotto:
File Dimensione Formato  
jnl-2013-tcpmt-parallel.pdf

accesso aperto

Tipologia: 2. Post-print / Author's Accepted Manuscript
Licenza: PUBBLICO - Tutti i diritti riservati
Dimensione 718.22 kB
Formato Adobe PDF
718.22 kB Adobe PDF Visualizza/Apri
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11583/2510284
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo