PC microprocessors and digital signal processors (DSPs) are getting faster and more complex every day. This makes the task of the power supply designer much more difficult. It also places a great deal of importance on simulating the operation of a point-of-load processor power supply to test its static and dynamic voltage regulation, efficiency, and thermal performance long before it is implemented in a production system.
Many processors now have hundreds of millions of transistors, run at frequencies approaching a gigahertz, and as a result, draw significantly more current than CPUs did just a few years ago. In fact, it is not uncommon for a PC processor to draw 20 to 50 A; a desktop system of only five years ago drew approximately 10 A.
But while the current requirements of processors increase, core voltages are decreasing to less than 2 V and, in some cases, less than 1 V. At the same time, the allowable voltage variation for processors is falling with the supply voltage.
An allowable 500-mV variation on a 5-V system becomes less than 100 mV for a 1-V system. Coupled with almost an order of magnitude current increase, power supply load transient response and power distribution design become paramount concerns and can drive up the cost of power supply components, especially capacitors.
Higher currents also can degrade power supply efficiency, causing the power supply to dissipate more heat and create thermal management problems. At the same time, motherboards are getting progressively smaller, which puts the heat-dissipating components closer together and further complicates the issue. The designer must use a wide variety of test equipment and methods to verify that the power supply design meets the severe requirements of voltage regulation, efficiency, and thermal performance.
Power Supply Controllers
Various kinds of synchronous regulator control techniques have been used in recent years, including fixed-frequency voltage-mode, fixed-frequency current-mode, variable-frequency current-mode, variable on-time only, and variable off-time only. All of these topologies have slow transient response times, requiring additional bulk capacitors on the output to maintain the voltage within the regulation limits. Capacitors add cost and take up space at a time when system manufacturers are desperately trying to reduce both of these parameters to remain competitive in the marketplace.
In contrast to these power supply controller topologies, hysteretic controllers featuring a fast comparator with a low hysteresis window now provide a fast transient response in the range of 400 ns to 500 ns as the load on the supply varies dynamically between no load and maximum load. With faster transient response times, fewer capacitors are needed, meaning that space and component costs can be reduced.
As with all power supply designs, the designer must carefully test the voltage variation at the processor during both static and dynamic conditions. These tests should be done under conditions that resemble as closely as possible the actual operating conditions of the processor.
Test Setup
To fully test the operational characteristics of a typical motherboard power supply, a test setup much like the one shown in Figure 1 is needed. This test bed or test fixture includes the prototype of a point-of-load power supply, processor bypass capacitors, and a transient load tester to manipulate the load on the power supply. In this example, the prototype supply design implements TI’s TPS5211 Hysteretic Power Controller in a synchronous buck converter architecture.
With this type of fixture, the designer can optimize the power supply design in terms of output inductor value, switching frequency of the converter, and the number of bulk capacitors that will be needed. In turn, these factors will affect the cost and size of the system.
A block diagram of the synchronous buck converter with the hysteretic controller that was used on this test fixture is shown in Figure 2.
Test Results
Using the test bed shown in Figure 1, a number of tests were run to optimize the design of a power supply. In this case, test results were measured against the requirements of an Intel Celeron processor with a 733-MHz clock and 133-MHz bus frequency.1
The first test verified the steady-state output voltage. This ensured that a 1.52-V to 1.64-V range was maintained over a varied input-voltage and load range. In this example, the output voltage would depend on the load current because the active droop compensation has been used to improve the output-voltage transient tolerance and reduce the number of capacitors.
The droop compensation decreased the output voltage as load current increased to position the output voltage to take advantage of the full regulation window. Without droop compensation, during a transient, the output variation can only be plus or minus the regulation tolerance. With the droop compensation, the power supply can fully use the total regulation limit.
The second important characteristic test is the load-current transient response. Figure 3 shows the waveforms for the output-voltage transient response using a hysteretic controller. In this example, the transient lasted only 7 µs for step-up and 34 µs for step-down, well within the 100-µs recovery time requirement for the Celeron processor.
This test reveals the fast transient response of the hysteretic controller. Test results show that the delay is 340 ns during the load-current step-down and 500 ns during the step-up. Theoretically, a voltage-mode or current-mode controller starts reacting to transients only at the next switching period. To get the same reaction time, these controllers must run at a switching frequency greater than 2 MHz.
In contrast, the hysteretic controller responds almost instantaneously when the load-current transient occurs. It improves efficiency because it can run at a much lower frequency which reduces switching losses.
Testing for Power Losses and Thermal Performance
Providing power efficiently is extremely important in certain applications. Test-equipment accuracy is critical in these types of measurements because the result is related to each of four measurements: voltage in, voltage out, current in, and current out.
For example, 1% inaccuracy in each measurement can lead to a 90% efficient power supply being represented as 86% or 94%. While seemingly trivial, in this instance, this range can represent a more than 2½:1 variation in the power supply loss. Using the test fixture, the efficiency of a design was ascertained over the entire input-voltage and output-current range. Measurements were made after 12 hours of operation when the temperatures of the PCB and components stabilized.
In this example, tests show that the efficiency of the design is well within the limits of the requirements for an Intel Celeron processor. Specifically, efficiency at the 22-A end of the range was 84% and 53.8% at 0.5 A. The specification calls for efficiency levels of 80% and 40%, respectively.
Thermal performance has become increasingly critical in recent years as high-speed, high-current processors have been deployed in small enclosures with little or no airflow. Again, by using the test fixture, measurements showed that the temperature change of the components and the PCB would not affect the reliability or performance of most applications with a Celeron processor. Measurements were made at room temperature with natural convection cooling. The input voltage was 5 V with a 22-A load current.
Figure 4
is a thermal map of the power supply. The thermal camera not only shows measurements of components which were expected to be warm, but also detects those that the designer may have overlooked. These measurements are within the requirements of the specification and imply that the actual thermal performance would improve when the design is installed on a motherboard with a larger surface area to dissipate heat more effectively.
The Test Imperative
Losses such as reverse recovery require simultaneous current and voltage measurement which, until recently, could not be performed in real time. Most current probes have delays in excess of voltage probes. These delays have had to be manually deskewed from voltage measurements.
Many times, switching losses are determined by gate-drive characteristics which often are difficult to measure on switching power devices. Amplifiers with high common-mode rejection ratios and DC isolation are required for accurate measurements.
Finally, the start-up of a power supply has been difficult to test. Large amounts of memory in digital scopes and equipment to measure duty cycle have helped to ease this issue.
Accurate simulation and testing of point-of-load power supply designs will be critical in the near-term as consumers and users demand systems that are more versatile, graphical, user friendly, and powerful. Test fixtures or test beds are critical to simulating the actual operating conditions so that a design can be tested over the system’s foreseeable operating range.
Reference
1. VRM 8.4 DC-DC Converter Design Guidelines, Intel.
About the Authors
Rais Miftakhutdinov is the senior systems designer for Power Management Products at Texas Instruments. Before joining TI two years ago, he had worked 20 years as a designer of power electronics for Russian and U.S. companies. Dr. Miftakhutdinov received a Ph.D. in electrical engineering from Moscow State Aviation Institute in 1997. MS 8710, (214) 480-2401, e-mail: [email protected].
Robert Kollman is the power management applications manager at Texas Instruments. During his 25-year career, Mr. Kollman has been employed at TI for more than 15 years and at Raytheon, MPSI, and Rockwell. He has a B.S.E.E. from Texas A&M University and an M.S.E.E. from Southern Methodist University. MS 8710, (214) 480-6653, e-mail: [email protected].
Copyright 2000 Nelson Publishing Inc.
Texas Instruments, 12500 Texas Instruments Blvd., Dallas, TX 75243.
June 2000