Electronics Cooling Discussed in the second blog in the series
We focused on Metal Core Printed Circuit Board (MCPCB) designs and their solutions that SinkPAD™ makes as the focus of our first blog in this series. We continue the discussion by offering a further look into the white paper we will be releasing.
The work presents a novel approach for predicting temperature evolution in electronic devices subjected to transient heat sources. It is based on modeling dynamic behavior of a thermal system with an identified network and discretized time-constant spectrum. We revisit the model reduction by network identification (NID) and present an extension of a method to obtain time-constant spectrum of a thermal network response, based on analytical form of convolving functions, while providing new insights to limitations of the technique. We verify the model extraction procedure using analytical solution and demonstrate correct identification of known system poles and convergence of the extracted time constant spectrum to the limiting case. This can be very useful in electronics cooling and thermal management. Heat transfer is the domain of mechanical engineering, even when utilizing advanced mathematical methods described herein.
We then present IIR digital filters suited for run-time evaluation of convolution integral in discrete time-domain. A simple formulation of recursive digital filters makes the algorithm well-suited for run-time temperature predictions. The resulting recursive algorithm yields temperature calculation at a given time instant using very limited depth of recorded temperature history. A numerical model of semiconductor device is created to generate time-domain temperature responses to step-function power excitation; excellent accuracy of the filter output is confirmed when compared to simulations. It is necessary to predict the temperature in real time to actively manage electronics cooling.
Comparison with conventional integral-based convolution methods also indicates a dramatic improvement in computational efficiency compared to existing techniques. The achieved scaling is best possible, linear, with number of temperature evaluations, a feature enabled by the use of a Digital Signal Processing (DSP) technique. This improvement allows implementation of sophisticated runtime dynamic thermal management algorithms for all high-power architectures and expands the application range to embedded platforms for implementations in a pervasive computing environment.