Declining microprocessor operating voltages complicate the task of CPU-core power conversion. A core power supply must exhibit fast transient response times, high efficiency and low heat generation in the vicinity of the processor. Those factors will soon force a move away from one-step power conversion (from a battery or wall adapter to the processor) to a two-step conversion, wherein the CPU-core power is obtained from the 5-V supply.
While new to the portable arena, distributed power systems using 5 V as a bus voltage have been used in large systems for many years. Although it may not be absolutely necessary to adopt such an architecture in portables today, the clock is ticking for the old brute-force approach.
Consider the biggest argument against two-step conversion: the perceived drop in efficiency and attendant heat generation in the 5-V supply. Off-the-cuff calculations give a false impression that efficiency significantly decreases. On the contrary, accurate calculations of efficiency for two-step power conversion based on actual demo-board measurements show efficiency numbers within 1 percent of high-efficiency one-step converters.
On the other hand, many benefits result from two-step conversion: a more symmetrical transient response, lower heat generation in the vicinity of the processor and easy modification for lower processor voltages in the future. Peak currents taken from the battery are also reduced. That improves battery efficiency, which can often compensate for the slight difference in efficiency measured using laboratory power supplies. Consequently, battery life in a real notebook computer is virtually identical for one-step and two-step architectures.
The most prominent design problem lurks in the duty cycle for a step-down switching regulator, given by the ratio of VOUT to VIN. In one-step power conversion, the switch on-time must be very short because the step-down ratio is large. That yields a very fast inductor current ramp-up and a much slower current ramp-down. The inductor size must be large enough to keep the current under control during the ramp-up. That requires a larger inductor than is required for operation with a low input voltage. Fast current rise and slow current decay mean that the transient response of the regulator is good for load increases but poor for load decreases.
The lower, constant input voltage for a two-step conversion process not only yields a more symmetrical transient response but also eliminates the headaches associated with optimizing loop dynamics over widely varying battery and wall-adapter voltages.
Because the duty cycles are closer to 50 percent with two-step conversion, and because there is less switching loss, the switching frequency may also be increased. That allows smaller, lower-cost external components to be used and further aids the transient response.
To minimize the high current pc-board trace lengths, the core supply must be located near the processor. With a one-step converter, the power lost is significantly higher than for the second step of a two-step conversion. Switching regulators for converting high input to low output voltages rarely approach 90 percent efficiency. A properly designed 5-V-to-core-voltage converter can add up to 5 percentage points of efficiency, thereby minimizing heat generation near the processor.
A common mistake when computing the efficiency of a two-step power-conversion system is to simply multiply the efficiency of the first conversion by the efficiency of the second conversion. While expedient, that method does not reveal the overall system efficiency or the distribution of losses on the board. The correct approach to evaluate two-step power-conversion efficiency is to return to the definition of efficiency: Efficiency = total power out / (total power out + total power lost) x 100 percent.
The total-power-out term must include not only the power supplied to the CPU core but also the power supplied at each conversion from which the CPU-core voltage is derived. The total-power-lost term is the sum of the powers lost at each conversion and is calculated from the respective operating efficiencies.
For example, assume that worst-case current levels of 3 A at 5 V and 10 A at 1.5 V are required in a notebook system. It is readily apparent that the total power out must be 30 W (3 x 5 + 10 x 1.5). When comparing the power lost at each stage for one-step and two-step conversions from a 15-V input voltage, we find the numbers are very close.
A total of 3.69 W of power will be lost in a two-step conversion approach, compared with 3.23 W for one-step conversion. Entering those numbers into the efficiency equation reveals only a 1.2 percent efficiency difference between one-step and two-step conversion. When other power losses in the 3.3-V and backlight circuits are included, the difference drops to less than 1 percent.
What about the increased burden on the 5-V regulator? While the power lost in the 5-V supply does increase with two-step conversion, it is still less than is lost in the one-step CPU-core supply. Further, power lost in the core supply is in the worst possible thermal environment for a notebook computer-next to the processor. In this example, two-step conversion reduced the power dissipated in the vicinity of the CPU by more than 0.75 W.
An additional concern sometimes voiced by power-supply designers is the pitfalls of loading the output of one switching regulator with the input of another. In fact, the input current of a switching regulator is directly proportional to its output voltage and current and is inversely proportional to its input voltage.
That represents a benign load for an upstream switching regulator, and cascaded switching regulators have been used in a host of different power-distribution applications over the years. Today's desktop computers, for example, use precisely the same architecture as proposed here for portables.
As time goes forward, microprocessor fabrication lithography will continue to shrink and will thus mandate even lower CPU-core operating voltages and higher operating currents. Already on the horizon for portable systems are 1.1-V supplies and 15-A operating currents. Those demands will render one-step conversion approaches unworkable as a result of infinitesimal duty cycles and severely skewed transient behavior.