Comparing hot swap IC solutions in server power reporting (Part 2 of 2)

(Part 1 looked at aspects of the voltage- and current-measurement challenges of measuring power in this application, along with error sources and tradeoffs; you can read it here.)

Integrated hot swap and power measurement solution

Figure 3 shows the benefits gained by combining the following in an integrated approach:

1. Advanced hot swap circuitry with both current- and power-limiting features to limit MOSFET power dissipation, while providing robust protection against voltage spike and inrush current for the components downstream

2. 1% accuracy high-side current sensing with 25 mV full-scale range reduces both power dissipation at the board level and wasted overhead power delivery at the system level

3. 1% accuracy built-in voltage division and sensing enables internal calculation for voltage and current, to provide power telemetry and averaging, valuable parameters for data center utilization

4. Internal ADC with sample and hold method allows simultaneous sampling of voltage and current to provide the ultimate power measurement accuracy

Figure 3:  Designing server instrumentation is made simple

by the use of an integrated solution like the National LM25066.

This device provides all the functions needed for precision telemetry,

and combines that with an advanced hot swap that limits MOSFET power dissipation.

Sampling considerations

The discussion of voltage variation is an excellent place to bring up the subject of sampling. Since analog-to-digital converters (ADCs) are used to perform this measurement, this will be a sampled system. This leads to sample-timing considerations, since power is being measured, which is the product of the voltage and current at a given time.

Because cost is a factor, it is desirable to use a single ADC and multiplex these signals. But a conventional system as shown in Figure 4 cannot take these samples at exactly the same point in time. Figure 4 also depicts with dotted connection a simultaneous sample system. To take the sample, S1 and S2 are closed at the same instant storing their values on the sample/hold capacitors. The ADC can then convert each channel sequentially.

Figure 4: Server instrumentation uses a sampled system with an ADC.

The simplest form will not take voltage and current samples at exactly the same point in time.

Adding a sample and hold capability enables simultaneous sampling with a single ADC.

Figure 5 illustrates the importance of sample timing. In systems that can only sample one parameter at a time, the voltage sampled does not match the voltage present when the current was sampled. This can cause errors in power measurement: consider that 120 mV variation in voltage corresponds to 1% error, and such variations occur frequently.

Figure 5: Note that a system that samples current and voltage at separate points

could miss a variation in one or the other of the parameters,

as shown by the separate lines for voltage and current sampling.

Simultaneous sampling on the other hand will capture this, which translates to more accurate power measurement.

Power calculation

Many end users prefer to have power data readily available, to save them the burden of making the calculation. This means the instrumentation IC must perform the multiplication of voltage and current to provide the power calculation. (One consideration is the units that the power will be reported in, and the most convenient solution to this is to make a relative power calculation).

 If the current value (actually represented by a shunt voltage value in most ICs) is at full scale, and the voltage is at full scale, the multiplication of those two values yields a full-scale power register. It is then up to the user to assign the correct LSB values which, of course, is a function of the exact value of shunt resistor used.

This type of reporting is readily accommodated by the PMBus with the direct-reporting format, which allows for coefficients to convert the IC data into "real world" numbers. Certain coefficients will be fixed by the manufacturer, such as the voltage-measurement scaling which is provided by a fixed, internal voltage divider. The end user will have to calculate current coefficients, since these are proportional to the exact value of current shunt resistor used.

Averaging and power calculation

Current measurements are frequently noisy, and in a digital system averaging readily smoothes the noisy readings. However, the methodology of averaging plays right into the accuracy of the power measurement. In power measurement, it would seem that there are two ways to go about averaged measurements.

The first example of averaging might be to average a number of current readings, then average a number of voltage readings, then multiply those values to yield averaged power, Equation 3:

Note that, in any case, averaging effectively increases sampling time and burden on the microcontroller/process if averaging is done outside of the device, thus extending the amount of time between which new valid data becomes available.

Data storage and alarms

"Power capping" is one of the functions that can take advantage of alarm capabilities in server instrumentation. Furthermore, because the power-control functions such as hot swap are combined with the telemetry, limits can be set such that impending fault conditions can be used to notify the system.  Peak values can be stored, as well as values present at the time a fault occurs. This provides a "black box" function yielding data on system conditions at the instant of the fault.

Interface

The SMBus is ubiquitous in the data center, server, and communications infrastructure world as a means of communicating within the system. SMBus is built on, and generally compatible with the I²C bus. A recent development is the PMBus, a bus which uses the physical layer of the SMBus and defines a protocol for communication with power-management systems and integrated circuits. The PMBus is well suited as a means of communication with server telemetry instrumentation.

Possibly the most important improvement PMBus brings is packet error checking, and error checking method. Many users of I²C and SMBus have lamented the fact that they have no way of knowing that writes and reads have placed or yielded the correct data. Other benefits are specific commands for reading many of the alarms and data points, and as time goes on this will become even more defined for server instrumentation.

Figure 6: LM25066 Block diagram showing key attributes and capabilities of the LM25066

3. System architectural impact

Firmware architecture for the LM25066, Figure 6, can be similarly compared to solutions with I²C as well as SMBus designs, thereby minimizing the architectural impact in customer systems.  All of the hardware and bus protocol are identical with SMBus. PMBus adds specific command pointer locations to keep the commands standardized across various manufactures for the future platform flexibility.

BMC requirements are relaxed, as the BMC no longer has the burden to calculate and average power. Since power is calculated and averaged by the LM25066, the BMC could read the LM25066 at its leisure. The occasional burden of running a calibration routine is also eliminated.

4. Advanced protection/reliability capabilities

The LM25066 integrates the hot swap with this power-monitoring system which is a logical integration and a reduction in board area.  LM25066 has several MOSFET protection features as part of the hot swap functionality such as

• MOSFET power dissipation limiting
• MOSFET thermal protection
• MOSFET failure detection

Temperature reporting is another useful function of server instrumentation, and the LM25066 has provision to use a remote junction (a transistor) as a temperature sensor. One way to use this would be to mount it in close proximity to the MOSFET, to provide for thermal limiting of the MOSFET. Alternatively it can be used to report temperature at virtually any location for almost any purpose.

This is combined with MOSFET power dissipation limiting, where the LM25066 actually measures voltage drop across the MOSFET, current in the MOSFET, and uses an analog multiplier to calculate the power dissipation of the MOSFET. There is no better method of protection, as this provides the closest fit to MOSFET safe-operating-areas possible.  Smaller MOSFETs can be specified with this protection

Conclusion

The growing pain of increasing data-center capacity, while reducing its energy footprint, has been challenging for many server manufacturers.  Increasing processing power in the same amount of rack space poses a board-level power-density challenge.  Meanwhile, techniques such as power capping have been introduced to increase energy efficiency in data center, but the ability to measure the power level accurately remain the key factor.

The National LM25066 saves significant board space and cost compared to the discrete implementation while providing power-monitoring accuracy and capability that can off-load the on-board processing engine. The LM25066 provides this with a PMBus interface using an industry-standard command set for easy future compatibility, without major impact to the system architecture.  Robust MOSFET protection and monitoring further ensure the ability to provide the most economical yet reliable power control solution.

Reference

•Energy Logic: Reducing Data Center Energy Consumption by Creating Savings that Cascade Across Systems.  Emerson Network Power, 2009.

About the authors

Joy Taylor is a product marketing manager at National Semiconductor Corp., focused on power management products and technologies. Taylor has held various roles in applications and product marketing at National since 2003. She has a BSEE from San Jose State University and an MBA from Santa Clara University.

Jerry Steele is a strategic applications engineer at National Semiconductor’s Tucson Design Center, specializing in defining power management products. Steele has over 25 years of experience in the analog and mixed-signal industry, has authored several articles and co-authored four patents.