CPU Power Controller: Performance Comparison of Single-Edge and Dual-Edge Architectures

In terms of cost and performance, the development of motherboard power systems will always encounter many challenges, especially CPU main processor power supply (Vcore) control. In the past few years, current requirements have increased from 30A to 130A today to meet the peak current requirements of current systems. Corresponding to the increase in current levels is a very significant increase in slew rate. Current system requirements require a Vcore power system that can handle conversions above 300V/μs and up to 2A/ns. Meeting these requirements is very difficult, but even more difficult is how to meet the voltage accuracy requirements of the CPU. In fact, as current levels increase and switching speeds up, the CPU's power supply must not be affected by these changes. The final requirement, and perhaps the most difficult to achieve, is that the system must cost less than the previous generation.

The days of using single-phase switching regulators to meet these stringent requirements are long gone. In recent years, semiconductor suppliers have introduced many new solutions to the market. The addition of DAC-controlled output voltages and later the move to multiphase controllers were significant developments. Multiphase power systems meet the power requirements of today's CPUs up to 130A by dividing the 130A current into four phases, resulting in a more controllable 32A current level in each phase. These systems require sophisticated controllers to balance current sharing and transient response through three to four phases in desktop computer systems and up to six phases in server systems. Different phases are individually turned on as needed to provide the output voltage. Each phase current is asynchronous to the other phases to reduce input filtering requirements. The problem faced by system designers is that only one phase at a time can immediately respond to the transient event and provide the extra power required by the CPU. This eliminates the need for a voltage controller to provide the additional power required for first- and second-order transient events. The initial response is provided by the energy stored in the output capacitor, and the regulator then follows up to meet the power demand.

Figure 1: A dual-edge modulation architecture can turn the controller on and off as needed.

The controller architectures used today are usually either trailing edge or leading edge, and each architecture has its advantages and disadvantages. Controllers with trailing edge control architecture turn on at the beginning of each clock cycle. The controller is able to respond to any transient event that occurs while it is on, but it must wait until the next clock cycle to respond to a transient event that occurs while it is off. A controller using a leading-edge architecture turns off during a clock cycle, at which time it can respond to transient events that occur during off-time, but it must wait until the next clock cycle to respond to transient events that occur during on-time. In both architectures, a latch is typically placed at the PWM comparator output, which creates a single-cycle delay in response to a transient event.

It is generally believed that the next step in Vcore voltage power supply control requires digital control systems to overcome these problems. Digital power controllers are not constrained by the same circuitry found in analog controllers and can overcome many limitations. In fact, there are already vendors with digital control architectures on the market. While these controllers do offer high performance, their high cost and the bold changes required to implement the solution hinder market acceptance. And because power system designers have accumulated so much experience and knowledge in analog systems, they are resistant to radical changes without clear benefits.

CPU power system designers need an inexpensive intermediate solution that allows analog controllers to overcome the constraints of current architectures while delivering the benefits of digital architectures. ON Semiconductor has developed a novel dual-edge architecture that can achieve these goals. This control scheme combines the advantages of leading and trailing edge architectures while avoiding the disadvantages of both. The dual-edge architecture (Figure 1) is not limited by the clock period that determines when to turn on or off; the turn-on signal is determined by the error signal. Similarly, the error signal indicates when the controller is off. And because no latches are included in the architecture, another source of response latency is eliminated. This architecture, combined with fast output feedback, allows all phases to respond to transient events. All phases provide power to the CPU immediately, eliminating the need for system decoupling solutions because the power controller does many things that were not possible with previous architectures.

ON Semiconductor's new CPU multi-phase power controller NCP5381 adopts this new dual-edge architecture. A direct comparison of single-edge and dual-edge controllers in an existing motherboard (using a three-phase design with a single-edge controller) shows that dual-edge performs better than single-edge in terms of transient response. Comparing the conduction characteristics difference between single-edge and double-edge architecture chips when working, the results show that the double-edge structure chip NCP5381 enables all three phases to respond to transient events simultaneously, while the single-edge architecture cannot enable all three phases to provide power at the same time, at most at any time There are two phases overlapping, and it takes longer to stabilize the output voltage.

Motherboard power system designers cannot increase prices just because performance exceeds VRM specifications. Therefore, the designer will test the design and remove capacitors or replace them with smaller values ​​until the specifications are met. These are trade-offs that must be made with every product generation. Using a dual-edge controller reduces decoupling costs without sacrificing system performance. Comparing the amount of undervoltage and overshoot required by each controller before the correct output voltage, the dual-edge controller has much less undervoltage/overshoot and has also been shown to transition to the correct output voltage faster. Combining the superior transient response of a dual-edge controller with the output capacitance solution used in this single-edge controller design results in excellent performance. This suggests that in order to optimize dual-edge controller boards, system designers must remove capacitance to match the performance of single-edge controllers currently used on boards. CPU power decoupling solutions used on motherboards today vary by controller and motherboard manufacturer, but the above direct comparison shows that the dual-edge NCP5381 can reduce the number of capacitors required. Using NCP5381 can save the number and cost of capacitors.