The increasing demand of mobile computing and handheld devices are driving architects to move from discrete components to tightly integrated System on Chip (SoC). The back-end servers are also following the trend to perform faster computations to meet the increased data processing demands. Providing longer battery life and being environment friendly is part of this proliferation. This requires more sophisticated power management schemes. The voltage regulator is a very important part of the power management and the placement of voltage regulator plays an important role in improving the performance.
Figure 1 shows a typical voltage regulator. The “Power Supply Unit” is the supply to regulator. The power supply for servers, desktops and laptops is typically 12V and for mobile, it is 3.3V. A 1.0V regulator is generally used to power core digital circuits while IOs are normally 1.8V. The duty cycle of the “Gate Driver” decides the output voltage of switching regulator. To remove the ripple on output voltage rail, a low pass L-C filter is used. For powering the sensitive analog circuit, a separate LDO is used.
Typical Voltage Regulator
The motivation for placing the voltage regulator on die arises from following benefits:
- Granular and Faster Power Management: The on-die voltage regulator offers more control for shutting down the unused circuits and saving power. The time required to become active from a sleep state is in the order of microseconds (μs) for off-die regulators while it is in the order of nano seconds (ns) for the on-die regulator. This helps provide tighter control of power consumption that helps in increased battery life for handheld devices and lower cooling cost for servers.
- Reduction in I2 R Power Losses: A typical server processor takes 65W of power at full load. Delivering it at 1V requires 65A of current delivery that requires thick traces to keep the I2 R losses low. Supplying the same power at 2V reduces the current to half (32.5A) which helps in reducing the track widths for the same I2 R losses. The reduction in I2 R losses also helps in reducing the board area.
- Saving on PCB area: The area available on the PCB is very precious and reducing the PCB area results in smaller form-factor. Placing the voltage regulator on die removes the components associated with the regulator which helps in saving the area on the PCB and reducing the BOM.
So why do some designs still prefer the voltage regulator off-die as opposed to on-die? To answer this question, let us discuss the building blocks of a voltage regulator and the tradeoffs involved in keeping in on-die vs. off-die.
- Power Train: Power train offers two conduction paths for voltage regulation, one from VIN to filter and other from filter to Ground. In all the voltage regulators, the path from VIN to filter is a switch made from either a PFET or NFET. On the other hand, the path from filter to Ground is either a diode or NFET. Using FETs in both the paths offers better performance for voltage regulation. The first tradeoff comes from the voltage rating on the FET. Off-die FETs come in several flavors and can support voltages beyond 400V. In contrast, on-die FETs, in a typical CMOS process, can support up to 3.3V and in the case of processors, it can be only up to 2V.
The power dissipated from the power train is another tradeoff. The Off-die FETs have additional area for heat dissipation, but On-die FETs increase the overall die temperature and require additional design for thermal management. Having the power train on die offers higher switching frequency in hundreds of MHz. Higher switching frequency reduces the size of filter. Having the power train on die also doubles the number of power pins, since additional pins are required to connect the output of power train to the filter.
- Inductor: The inductor, along with capacitor, forms a low pass filter that is used to suppress the ripple voltage at the output. The inductor required for voltage regulator is normally large so these cannot be kept on the die. The off-die regulator uses SMD inductors available off-the-shelf. The on-die regulator allows a higher switching frequency that helps in reducing the size of the required inductor to achieve the same efficiency and can be realized using PCB traces. This helps in reducing the number of components on the PCB at the cost of higher DC resistance, that increases losses.
- Capacitor: The capacitor helps in reducing the ripple at the output of voltage regulator. For an off-die regulator, the bigger capacitor is required to suppress the ripple due to a lower switching frequency. Using on-die regulators with high frequency switching, the on-board capacitors can be made small or can be even eliminated. However, this occupies significant area on die and generally has higher ESR than off-die capacitors.
- Controller: The voltage regulation loop controller has very little impact on deciding the placement of the voltage regulator. In the case of an off-die regulator, a separate chip is used for the controller while in the case of an on-die regulator, the controller is part of control loop on the die. The advantage of an on-die controller is better control of the regulation loop and the controller can be realized in the digital domain, which can be reused when the technology changes.
At the end, the decision of placing the voltage regulator, on-die or off-die depends on the applications. When the load currents are small, the input voltage is high, slower wakeup and sleep are acceptable or granularity of power management is not critical, an off-die regulator is a good solution. For these reasons, LED lights are often powered by off-die regulators. In contrast, if the load currents are large, the input voltage is low, faster wakeup or more granularity is required for efficient operation; an on-die regulator becomes a better choice. A typical desktop/server processor is an example of on-die regulator that helps in improving power management.