Although power management is critical to the reliable operation of modern electronic systems, voltage regulators are perhaps the last remaining “blind spot” in today's systems, without the means for directly configuring or monitoring key power system operating parameters. Digitally programmable DC/DC converters have been available for many years, most notably in voltage regulator module (VRM) core power supplies with voltage identification (VID) output voltage control. But the ability to monitor operating status information directly from the voltage regulator, especially real-time currents, has been missing.
A principal benefit of digital power system management is reduced design cost and faster time-to-market. Complex multi-rail systems can be efficiently developed using a comprehensive development environment with an intuitive graphical user interface (GUI). Such systems also simplify in-circuit testing (ICT) and board debug by enabling changes via the GUI instead of soldering in “white wire” fixes. Another benefit is the potential to predict power system failures and enable preventive measures, thanks to the availability of real-time telemetry data. Perhaps most significantly, DC/DC converters with digital management functionality allow designers to develop “green” power systems that meet target performance (compute speed, data rate, etc.) with minimum energy usage at the point of load, board, rack and even installation levels, reducing infrastructure costs and the total cost of ownership over the life of the product.
NEED FOR ACCURATE LOSSLESS CURRENT MEASUREMENT
Advances in DC/DC converter design make significant efficiency improvements difficult, and the next generation of power systems will reduce the power consumption through system-level dynamic load balancing. The algorithms controlling point of load converters need high accuracy power consumption data in order to fine-tune the models and optimize power distribution. This is where the benefits of real-time telemetry, combined with high analog accuracy, become apparent. Measuring power consumption with high precision requires a known resistive element to measure the output current. While a calibrated external shunt is accurate, it introduces additional power loss and makes the converters more expensive. A lossless alternative is to measure the average voltage drop across the parasitic DC resistance of the inductor (DCR), which saves component count and simplifies board layout. Compared to other lossless measurement techniques such as SenseFET, DCR current measurement is more cost-effective and simpler to use.
One significant drawback of the DCR method is the strong temperature dependence of the inductor resistance and the difficulty in measuring the exact inductor core temperature. Without proper temperature compensation, a change in inductor temperature of only 1°C corresponds to approximately 0.39% current measurement error. To make matters worse, inductor self-heating of tens of degrees can be observed at high loading levels. The algorithm described next (patent pending) can compensate for all these effects, and when calibrated, achieves accuracy better than ±0.25% across the full temperature and load current range.
DCR TEMPERATURE COMPENSATION
Placing a temperature sensor in close proximity to the inductor provides first-order temperature compensation. The accuracy of temperature compensation is increased if the temperature sensor is away from other significant heat sources, such as the power FETs. Heat dissipation in the inductor under high load conditions creates transient and steady state thermal gradients between the inductor and the temperature sensor, and the sensed temperature does not accurately represent the inductor core temperature. This temperature gradient is clearly visible in Fig. 1a, which shows a thermal image of the integrated DC/DC converter LTC3601 providing 1.8V, 1.5A to the output load. Fig. 1b shows the board layout.
In addition, transient heating/cooling effects have to be accounted for in order to reduce the transient errors introduced when load current changes are faster than the heat transfer time constants of the inductor. Both of these problems are addressed by introducing two additional parameters: the thermal resistance θis from the inductor core to the on-board temperature sensor, and the inductor thermal time constant, τ. The thermal resistance θis (°C/W) is used to calculate the steady state difference between the sensed temperature TS and the internal inductor temperature Ti for a given power dissipated in the inductor Pi
The additional temperature rise is used for a more accurate estimate of the inductor DC resistance Ri
Vdcr = Inductor DC voltage drop
Iout = RMS value of the output current
RO = Inductor DC resistance at the reference temperature, Tref
α = Temperature coefficient of the resistance.
Most inductors are made of copper, so we can expect a temperature coefficient close to αCU = 3900 ppm/°C. For a given α the remaining parameters θis and RO can be calibrated at a single temperature using only two load currents:
Rk = Vdcr,k/ Iout,k = Inductor resistance
Pk= Vdcr,k × Iout,k = Power dissipation
Tk= Sensed temperature, K = 1,2
These values are recorded for each load current. To increase the accuracy in calculating θis, the two load currents should be chosen around I1 = 10% and I2 = 90% of the current range of the system.
The inductor thermal time constant τ models the first order thermal response of the inductor and allows accurate DCR compensation during load transients. During a transition from low to high load current, the inductor resistance increases due to self-heating. If we apply a single load step from the low current I1 to the higher current I2, the voltage across the inductor will change instantaneously from I1R1 to I2R1 and then slowly approach I2R2. Here, R1 is the steady- state resistance at the given temperature and load current I1, and R2 is the slightly higher DC resistance at I2, due to the inductor self-heating. Note that the electrical time constant τel = L/R is several orders of magnitude shorter than the thermal one, and “instantaneous” is relative to the thermal time constant. The two settled regions give us the data sets (I1, T1, R1, P1) and (I2, T2, R2, P2) and the two-point calibration technique (Equations 3 and 4) is used to extract the steady-state parameters θis and RO (given a previously characterized ‘average’ α). The relative current error calculated using the steady-state expression (2) will peak immediately after the load step, and then decay to zero with the inductor thermal time constant, τ.
The time constant τ is calculated from the slope of the best-fit line
y = In (ΔI╚I) = a1 + a2t,
In summary, a single load current step is all that is needed to calibrate the DCR current measurement. The stable portions of the response give us the thermal resistance θis and nominal DC resistance RO and the settling characteristic is used to measure the inductor time constant τ. Once the self-heating parameters θis and τ are characterized once for a particular system, only the inductor DC resistance RO needs to be calibrated to compensate for the inductor tolerance.
HIGH PRECISION DIGITAL TELEMETRY
Linear Technology's new LTC2974 is a quad system supervisor that incorporates a 16-bit data acquisition system with best in class ±0.25% total unadjusted error. It provides digital read back of output voltages, currents and temperature on all four monitored DC/DC channels, using the PMBus interface. The current measurement uses the described DCR temperature compensation algorithm, and achieves order of magnitude higher accuracy compared to the prior solutions (Fig. 2). The LTC2974 enables great system flexibility, based on the device's combination of telemetry with precision fast hardware supervision, trim DACs and comprehensive set of sequencing and tracking options. The device's onboard EEPROM enables completely autonomous operation, without the need to develop software. The LTC2974 includes a fault logging capability via an interrupt flag along with a “black box” recorder that stores the state of the converter operating conditions just prior to a fault.