Measuring Power and Energy of CMOS Circuits: A Comparative Analysis J. Rius, A. Peidro, S. Manich, R. Rodriguez Departament d'Enginyeria Electrònica, Universitat Politècnica de Catalunya Diagonal 647, 9th floor, 08028 Barcelona rius@eel.upc.es E. Boemo Escuela Politécnica Superior, Universidad Autónoma de Madrid Ctra. de Colmenar Km. 15, 28049 Madrid eduardo.boemo@uam.es Abstract--- This paper presents and compares a set of experimental results on the measurement of power and energy consumed using an ammeter and a non-conventional capacitor-based method. A standard cell CMOS multiplier is utilized as a circuit under test. The paper shows, as main result, that the capacitor-based method allows the designer to obtain information unreachable with conventional ammeter measurements. I. INTRODUCTION Power and energy consumption are critical issues in the design of CMOS digital circuits and systems and a great amount of effort has been devoted to estimate these parameters in the early design phases [1][2]. Nowadays, a set of useful tools is being marketed to know in advance the amount of the power consumption of the fabricated circuits. However, actual measurements of power are still mandatory. First, during the manufacture and testing stages, it is necessary to know the real power consumption to check if the circuit fits the specifications. On the other hand, at system level or in circuits where the power consumption depends to a large extent on the data processed, it is easier and convenient to perform measurements to know the power/energy consumed by such circuits. The standard procedure to measure power/energy is to connect an ammeter in series with the power supply pins, as is reported in [12][13][7][8][9]. This is a simple technique, but the amount of information that can be extracted from this measurement is limited. Thus, other methods have been proposed. In [10], are presented a summary of several of such techniques. Jenkins et al. in [5] uses a small resistor (2Ω) in series with the supply pin, and measure with an electron-beam probe the drop voltage in the resistor. The same approach is used by Callaway et al. in [11], but using a 100Ω resistor. In [3], the author also uses a resistor and give several advises on both the technique and instrumentation. In [15], the drop voltage in a resistor is utilized to determine if the transient current is abnormal or not. Thus, possible defects in the circuit can be detected. Other approach is inspired by the Keating-Meyer technique for I DDQ testing [18] and by the MTCMOS solution to leakage power control [6]. This technique (hereafter referred as C-based measurement) uses a switch and a capacitor in parallel with the supply pins to measure the energy consumed in a given period of time [17]. In [4][16] the authors present a variation of such technique that uses four switches and two identical capacitors to perform the measurement in synchronous digital ICs in a cycle-by- cycle basis. The information published, however, does not deal with the problems related to the switch charge injection and capacitor matching, and only relative energy measurements are presented. The goal of this paper is to check and compare by experiment the capabilities of two measurement techniques: ammeter and C-based methods. Several experiments have been made by using a cell-based CMOS integrated circuit with different logic depth as a benchmark circuit. The rest of the paper is organized as follows: the next section summarizes the measurement methods. Section III describes the results obtained from measurements made on a board containing a full custom multiplier. Finally, in section IV the conclusions of the work are presented. II. METHODS TO MEASURE THE CONSUMPTION OF CMOS CIRCUITS A. General By measuring the instantaneous current supplied to the IC, i DD (t), it is possible to calculate the charge, energy and power consumed by an IC. The main problem lays on the proper measurement of i DD (t), which is difficult in most cases because of the i DD (t) waveform characteristics. That