Power Estimation Tool for Aurix 3g Microcontroller

Abstract

An advanced microcontroller can have numerous configurations based on the applications and its power requirement depends on such configurations. This paper proposed a methodology to estimate the power consumption of microcontroller application using the Machine Learning models. The model is developed from device configuration and measurement data captured from the physical device. Two ways of model development approachFlat and Hierarchical, are presented in this paper. With flat model approach the error % is nearly around ±10% and with hierarchical model, the error % is around ±5%. Estimating the energy consumption of applications is a key aspect in optimizing automotive microcontroller embedded systems energy consumption. Power constraints are increasingly becoming the critical component of the design specification of these systems. A new approach for power analysis of automotive microcontroller is being proposed. The idea is to look at the power consumption in an automotive microcontroller from the point of view of the actual software or instructions executing on the processor. The basic component of this approach is a measurement based, instruction-level power analysis technique. The technique allows for the development of an instruction-level power model for the given processor, which can be used to evaluate software in terms of the power consumption, and for exploring the optimization of software for lower power.

Existing methodologies are using architectural level analysis of microprocessors, in which the power cost of a module is given by the estimated average capacitance that would switch when the given module is activated. Since the above technique works at higher levels of abstraction, the power estimates they provide are not very accurate. And also modules like GTM (Generic Timer Module), SPU (Signal Processing Unit) have power dependency based on the register configurations and the current behavior of them were as expected. But Tricore, has different input parameters and also the parameters are less as compared to other modules and since the clock modification for the tricore cannot be changed independently and the difficulties in changing the tricore clock made the current behavior of tricore power estimation model not as expected. For example, current should increase with increase in IPC (Instructions per Cycle) but beyond 2 IPC the condition fails. Hence, the aim of this project is to create an instruction-level measurement based tricore model that includes all the instruction set, different kind of memories, data and also the operands which can be used to evaluate software or the instructions that are executing in terms of the power consumption in pre-silicon environment.