Two of the most recent trends in high performance DPA (Distributed Power Architecture) systems are the need for extremely fast transient response and the usage of IBA (Intermediate Bus Architecture). The need for fast transient response from the power system is due to the dynamic current requirements of the latest high speed circuit chips used in telecom and datacom applications. The IBA approach places inexpensive non-isolated converters in close proximity to the circuitry they power to address the transient current requirements. These POL converters are in turn powered by an isolated bus converter which transforms the system's input power to a voltage level that can be safely and efficiently utilized by the POL devices " often in the vicinity of 12V, 5V or 3.3V. Most POL converters operate in conjunction with external capacitance on their output to achieve the required transient response performance. This capacitance typically takes the form of several solid tantalum and MLC capacitors to address both bulk energy storage and high frequency decoupling needs. We will see that the capacitors needed for such a conventional system will, in aggregate, have a significant negative impact on the power system's cost, reliability and required PCB area.
Some point of load converters, such as a family of POL converters by Artesyn Technologies, are designed to operate without external bulk energy storage capacitors, providing significant advantages for the designers of IBA power systems. Not only do these converters eliminate the need for many output capacitors, they offer extremely fast transient response characteristics. These new POL converters are packaged in industry standard 1.30 x 0.53 inch footprint surface mount chips. They are capable of 15 amps of output current, are powered from a nominal 12V intermediate bus voltage, and are intended for a wide range of telecom and datacom applications. For example, the SMT15F-12 series currently consists of four models with nominal output voltages of 1.0, 1.2, 1.5 and 1.8V.
The 1.2V converter has a maximum output voltage deviation of only 43mV during a 5 amp output current change at a ramp rate of 100A/μs. The recovery time is less than 10μs. These fast current transitions are difficult to achieve with conventional electronic loads, requiring a special load card to enable efficient testing and characterization of converters under stringent dynamic conditions.
Figure 1 shows the typical transient response of this type converter. It is achieved with a high converter operating frequency (1MHz) that is designed around a monolithic silicon controller that integrates four active components of a buck converter-switching regulator on one piece of silicon. The IC contains a high-frequency PWM (pulse width modulator) controller and high-current gate drive circuitry optimized for switching two high-performance MOSFETs used for high-side control and low-side synchronous switching of the output. This means that most bulk and mid-frequency energy storage and decoupling capacitors can be eliminated from the power system design. The only required capacitance is the small value MLC devices that the load IC manufacturer requires to be located adjacent to the IC pins.
Figure 1--New PoL Transient Response Characteristics - Vout 1.2V
Solving the Capacitor Problem
The benefit of eliminating output capacitors by using the SMT15F can be substantial, due to the large number of capacitors used in the power distribution networks of the latest electronics. Figure 2 shows the capacitor count by value in an Intel Pentium 4 personal computer chip set. Figure 3 shows similar data for various types of printed circuit boards manufactured by Nortel Networks. Each of these hundreds of capacitors contributes to system cost and occupies valuable PCB real estate. Each capacitor also adds failure rate to the power system, not only for the capacitor's intrinsic failure rate, but also due to the two additional solder interconnects per capacitor. So, clearly, it is advantageous to eliminate as many of them as possible.
Figure 2--Capacitor Usage on a Typical Personal Computer Motherboard
Pentium 4 Motherboard, 845 GE/PE type Chipset Source: Murata Technical Paper
Some of the small value capacitors will need to remain adjacent to the load IC pins to satisfy the very high frequency transient demands of the circuit switching and the physical placement specifications of the IC manufacturer. However, most of the mid-frequency and bulk capacitors are candidates for elimination when using a fast transient response PoL converter in a properly designed power distribution system. A recent study by Xilinx (reference 4) indicates that low ESL decoupling capacitors with values over 0.01μF are still effective when mounted several inches away from the load as long as low inductance pad and via designs are used to interconnect the capacitor to the power planes. This implies that a fast transient response PoL converter using SMT interconnects on a reasonably sized PCB can eliminate all but the smallest value capacitors in many systems. To be conservative, we will assume that only capacitors with values over 0.1μF and less than 1000μF are candidates for elimination. Even so, the data in Figures 2 and 3 indicate that over a hundred capacitors could be eliminated on many PCBs if power converters with fast transient response capability were used.
Source: Nortel Networks Technical Paper
Figure 3 Capacitor Usage on Typical PCBs
Candidates for elimination with fast transient response PoL include 125, 130 and 146
Assuming an average cost of $0.13 each per capacitor and the elimination of 130 capacitors per PCB, the data in Figure 3 indicate that cost savings of about $17 per board is possible due to the elimination of capacitors when using fast transient response PoL converters. Along with this reduced cost come the added benefits of reliability improvement and a saving of valuable PCB real estate. Assuming an average failure rate of 9 FIT per capacitor, the average PCB could see a 1200 FIT improvement in failure rate just due to the elimination of capacitors.
Comparison with Conventional POL Converter
This type converter could represent a breakthrough in transient response performance. Figure 4 compares its transient response specification with that of a popular competitive PoL converter. Both converters are manufactured in a 1.3 inch by 0.53 inch footprint SMT package and are capable of 15A of output current at 1.2V with a 12V input voltage. The new type converter operates at three times the frequency of the typical converter. This, along with its proprietary control circuitry, is a big factor in achieving the specified transient response performance.
One of the biggest differences between the two PoLs is the way the transient response is specified by the manufacturer. The typical product is specified at a current ramp rate of 2.5A/μs (3.2μs for the 8A load change). The new converter uses a ramp rate of 100A/μs.
Features Common to Both Converters:
15A Output Current Capability
12V Nominal Input Voltage
1.3 x 0.53 inch SMT Footprint
Figure 4--Comparison of the new converter to the typical PoL converter
This exceedingly fast ramp is used as an acknowledgement that today's circuitry generates very fast transient currents that need to be accommodated by the power system. It uses a 5A current transient in the specification, so the entire current transition occurs in only 50ns. In spite of this stringent test condition, it has a voltage dip of only 43mV for this transient test condition. The voltage deviation will be roughly linear with the current step, so that the expected voltage deviation using a current step of 8A, would be about 69mV. This is significantly less than the 200mV specification of the typical product.
The recovery or settling time is also important both as an indicator of the speed of the PoL, and as a predictor of performance in actual systems. Since decoupling capacitance must be used to supply energy to the load until the completion of the settling time, the settling time, along with the voltage deviation of the PoL, will determine the amount of energy needed from the decoupling system and the number and type of decoupling capacitors required. The new converter has a recovery time of 10μs vs. 25μs for the competitive device.
It is important to note that the converter specifications are valid with no external capacitance on the converter. The typical PoL requires a 10μF tantalum and a 1μF MLC capacitor on the output to meet its specifications. In addition, the typical converter will probably need more external capacitance in the form of low-frequency and mid-frequency decoupling in order to meet the requirements of most power systems.
Figure 5--Dynamic Response Comparison
This can be understood more clearly by superimposing the transient response curves of the two PoL converters as shown in Figure 5. The area under their respective response curves represents energy that must be supplied by the power system decoupling network if the depression in the output voltage is to be minimized. As can be graphically seen in the figure, the areas are vastly different. The new PoL converter requires only small high-frequency MLC capacitors at the load device pins. The typical PoL will need a significant amount of external capacitance if the voltage excursion down to 1.0V must be eliminated. For example, to limit the voltage deviation to 100mV rather than 200mV, the needed capacitance can be roughly estimated by the relationship:
C = I dt/dv
With I = 8A, dt = 25μs and dv = 100mV,
C = 2000μF
In practice, the external capacitance would be composed of several smaller valued capacitors to take advantage of the reduced ESL and ESR of paralleled devices. These capacitors should not be needed when using the newer PoL converters such as the Artesyn SMT15F.
Figure 6 shows a typical system application. In this example, three FPGA chips are powered by a single PoL converter. Each FPGA can demand up to 5A of peak current, so a 15A PoL is used. The decoupling arrangement shown is as suggested by the FPGA supplier. Several (approximately 30 to 40) small MLC capacitors are required to be mounted adjacent to the array pins on each FPGA. These capacitors are 0.1μF or less in size. The suggested mid frequency decoupling is an array of six 47μF tantalum chips per device. These do not need to be located in proximity to the array pins. A 1000μF capacitor, either aluminum or tantalum, is suggested for low frequency decoupling, and its location is not at all critical.
Figure 6--Application Example using PoL Converter
Due to the fast transient response capability of the new PoL, these mid frequency and low frequency decoupling capacitors should not be required. As long as sound layout guidelines are used with the power and ground planes and low inductance interconnects used between the PoL and the PCB, the new PoL and the small MLC capacitors should satisfy the circuit's transient demands.
In this example, the elimination of the capacitors when using the new PoL provides several advantages, as shown in Figures 7a and 7b. The cost savings associated with the capacitor removal and the savings in PCB real estate more than offset the slight price premium for this high performance PoL. Additional system benefits accrue as system reliability is also enhanced by the removal of the failure rates associated with the decoupling capacitors and their solder interconnects and the free PCB space can be utilized by for other functions.
Cost Comparison - Large key Account
Additional Disadvantages of Competitive PoL:
PCB Area for Caps and mounting Pads1 764mm2
Failure rate adder for Caps2 402 FIT
Figure 7a Comparison of PoLs in Application Example
Cost Comparison - Smaller Account
Additional Disadvantages of Competitive PoL:
PCB Area for Caps and mounting Pads1 -- 764 mm2
Failure rate adder for Caps2 -- 402 FIT
Figure 7b -- Comparison of PoLs in Application Example
1. 47μF 6 mm x 4 mm, 1000μF 8mm x 6mm.
2. Failure rate 9 FIT each for Tantalum, 240 FIT each for Aluminum.
The additional disadvantages of the competitive PoL are:
1. PCB area for Caps and mounting Pads 764 mm2
2. Failure rate adder for Caps 402 FIT
Finally its is also worth noting that with a turn on time of 5ms, the new PoL is particularly suited to powering this type of FPGA, whose typical turn on time range coincides with the new PoL turn-on.
Loads with large transient content are becoming quite common and require new solutions from the power system if they are to be dealt with economically. The intent here was to acquaint you with the new PoL converters and show how their transient response will redefine how power systems are designed. Not only do they meet the transient requirements of the most demanding circuitry, but they do so while allowing the elimination of many decoupling capacitors. The net benefit is reduction in capacitor costs, system failure rate and component PCB footprint.
1) S. Kiriyama, Murata Mfg. Co., "Monolithic Ceramic Capacitors Help Reduce Noise in Personal Computers", (AEI, Dempa Publications, March 2003).
2) Nortel Networks, "Power Decoupling Capacitor Application Trends and Comparison with Future Embedded Capacitor Core", (Advanced Embedded Passives Technology Consortium, Jan 2003).
3) P. Markowski and C. Quinn, Artesyn Technologies, "Capacitor Selection and Placement for Fast-transient Point-of-Load", (Power Electronics Technology, June 2003).
4) M. Alexander, Xilinx, "Power Distribution System (PDS) Design: Using Bypass/Decoupling Capactiors", (Xilinx Ap. Note XAPP623, June 2003).
5) M. Alexander and A. Lesea, Xilinx, "Powering Xilinx FPGAs", (Xilinx Ap. Note XAPP158, Aug 2002).