Specification and Validation for Heterogeneous MP-SoCs

Transcription

1 Specification and Validation for Heterogeneous MP-SoCs Gabriela Nicolescu Ecole Polytechnique de Montréal Tel : (514) ext 5434 Fax: (514) gabriela.nicolescu@polymtl.ca

2 Heterogeneous SoC MEMS RISC Hw Components Config. Proc. Optical components Electro- Biological Components Communication Optical interconnect interconnect Paradigm shift SoCs are drivers for several technologies integration Applications automotive, communications, medical, defense 2

3 New technologies for heterogeneous SoC 3D System In Package Integration Specific components are fabricated on individual wafers and then integrated onto a single chip- scaled package Benefits Increased performance Increased integration density Reduced power consumption Wireless communication schemes Based on Capacitive coupling Based on Inductive coupling New system-level trade-off Source : Balinga, Banerjee 3

4 Outlook for the design of heterogeneous SoC Access to physical prototyping for multi- technology SoCs is a major challenge Significant cost Harder to influence standard processes Modeling and simulation becomes a necessary alternative in design space exploration for these systems Few existing approaches More research needed 4

5 Heterogeneous SoC Specification & Validation Optical Control Mechanical Extensions of existing tools/languages Homogeneous environment Classical HDLs + AMS concepts + new features for sim scheduler VHDL-AMS, Verilog- AMS, SystemC-AMS No powerful libraries 5

6 Heterogeneous SoC Specification & Validation Optical -1e-005-8e-006-6e-006-4e-006-2e e e-006 2e-006 4e-006-6e-006 6e-006-8e-006 8e-006 1e-005-1e-005 "chatoyant/wf4.txt" 1e-005 8e-006 6e-006 4e-006 2e-006 Control Electromechanical MoC Interfaces Heterogeneous Models of Computation (MoC) Single formalism for representing different models Deep conceptual understanding Ptolemy [Lee], Rugby [Jantsch] 6

7 Optical Heterogeneous SoC Specification & Validation Control Mechanical Heterogeneous execution models Multiple environments Taking into account implementation aspects Application specific efficient libraries LEOM [O Connors[ Connors] Pittsburgh [Levitan[ Levitan] Global execution models TIMA [Jerraya & Kriaa] Ecole Polytech Montreal Interfaces for Execution Models Adaptation 7

8 e-006 "chatoyant/w f4.txt" -4e-006-2e-006 1e-005 6e-006 8e-006 Key Features for Next Specification & Validation Tools Homogeneous environment facilitating cooperation between different teams Enabling easy specification, automatic generation for simulation interfaces Taking into account implementation choices Exploiting powerful existing tools (Simulink( Simulink,, SystemC, ) Based on a single well defined formalism for domain interaction Control Optical -1e-005-8e-006-6e-006-4e-006-2e e-0064e-0066e-0068e-0061e-005-1e-005-8e e-006 4e-006 Mechanical 8

9 Heterogeneity example - Continuous vs. Discret - Control t Discrete Model Mechanical t Continuous Model Simulation Step Synchronization State Event 9

10 Challenges for Accurate Global Execution Detection of state events generated by the continuous simulator Detection of the next event of the discrete simulator (scheduled event) Detection of the end of the discrete simulation cycle and the time step sending 10

11 Generic Architecture for Continuous/Discrete Simulation Discrete model Discrete simulator Continuous model Continuous simulator Détection d'événements Détection fin du cycle et Synchronization layer Synchronization Détection d'événements layer Cosimulation envoi du temps Changement de contexte interface Communication layer Communication layer Cosimulation bus 11

12 Generic Architecture for Continuous/Discrete Simulation Discrete State events detection End of discrete sim. cycle detection Context switch Communication layer Continuous Indication of state events and time sending Events detection Context switch Communication layer Cosimulation interface Cosimulation bus 12

13 First results - Continuous/Discrete simulation - SystemC/Simulink accurate simulation Easy integration, generic library elements Systemc Simulink 13

14 Performances analysis Inter-Simulators Communication overhead 20% of the total simulation time Overhead caused by the Simulink integration step adjustment max. 5% of total simulation time SystemC Synchronization overhead max. 0.2% of the total simulation time 14

15 Conclusions SoC drivers for multi-technology integration New CAD tools for design exploration are required Global specification and validation are important challenges First prototype for electro-mechanical systems Continuous/discrete integration Simulink and SystemC integration 15