Miniaturisation of VLSI (Very Large Scale Integration) devices has allowed for their usage in versatile applications such as baby monitors and heart rate monitors. One drawback to such miniaturisation of integrated circuits is the reduced capacity and lack of accessibility of the batteries used as power sources. This requires that the power consumed by the device be reduced as much as possible. To address this problem, Sagar Gubbi and Bharadwaj Amrutur of the department of Electrical and Communication Engineering at Indian Institute of Science (IISc) have developed a novel method which manages power better. The new method more efficient, reliable, and produces a hundred-fold improvement for certain operating conditions.
The current consumed by an electronic circuit depends on the voltage supplied to it, and hence the total energy consumption tends to decrease with power supply voltage. This happens until a point when leakages, i.e. energy losses, which occur even when there is no circuit activity, start to cause it to rise. Circuits are designed to operate at this point, called the minimum energy point. As Sagar elaborates, “the minimum energy point depends on the ambient temperature, varies from one chip to another due to process variations, and is also dependent on the [device] activity...So, it is desirable to track the minimum energy point during operation.” In their paper, Sagar and Amrutur propose an entirely digital circuit, a tiny square of side 30 microns, to achieve this.
Existing methods in literature estimate the minimum energy point by powering off a circuit and measuring the decrease in its input voltage. The novel all-digital method, however, does not rely on analog voltage measurements. This optimising circuit can be integrated into a circuit which forms a part of a miniature, ultra-low power device, to “control supply voltage to parts of a bigger system like a mobile phone”, explains Sagar.
Simulations performed and reported by the authors readily show the advantages of this procedure. The circuit can consistently estimate the minimum energy point to an error of about 5%, while existing analog methods have varying errors of up to 500%. Indeed, as Sagar says of his paper, “just as software eats the world, digital circuits are eating analog circuits”.
About the authors: Sagar Gubbi is a PhD student and Bharadwaj Amrutur is an Associate Professor in the Department of Electrical and Communication Engineering at Indian Institute of Science, Bangalore. Sagar Gubbi is one of his PhD students.
About the paper:The paper appeared in IEEE Transactions On Very Large Scale Integration (VLSI) Systems. It can be accessed at http://chips.ece.iisc.ernet.in/images/7/75/Sagar_Min_Energy_TVLSI.pdf.