The need for power-interface standardization
The expansion of computer functionality through add-on peripherals requires interface standards in order to realize the full potential of various vendors’ applications. Increasing the functionality of a desktop by adding wireless communication, or increasing the capacity of a laptop computer by adding more memory, allows a low-cost entry-level computer to be upgraded or tailored to an individual’s needs.
The early 1990s saw the advent of standards for PC add-on cards that allowed external memory from disparate vendors to be added to laptops. The PCMCIA (Personal Computer Memory Card International Association) was formed to standardize the interface, allowing the expansion of laptop memory using a flash or hard disk drive in the form of plug-in add-on cards. Not surprisingly, numerous other vendors quickly realized that their own specialized functions could also be added through PCMCIA cards.
Manufacturers of storage, communications and gaming applications, to name a few, joined PCMCIA in order to understand the interface or to influence the standards that would allow laptops to use their devices. With the diversity of host systems and card applications, it soon became apparent that the operating- and startup-power requirements of the cards needed careful consideration to prevent power-delivery and system malfunctions.
For example, the disk-drive motor startup or power hold-up capacitors required by many applications were a potential problem. They could cause massive in-rush currents that would overburden the host's power supply, causing system crashes or exceeding the Safe Operating Area (SOA) of the host-delivery MOSFET power switch. Voltages, currents (including surge currents) and sequencing were among the issues addressed by the PCMCIA standards committee. Although the PCMCIA has long been disbanded, its legacy of standardizing power delivery specifications now applies to various other add-ons, including the PC Cards that superseded PCMCIA cards.
System design approach
Like a PC Card, PCI Express (PCIe) addresses power requirements for add-on cards in PCs. The same power-delivery considerations apply and, like PC Cards, PCIe cards can generate secondary voltages that, depending on the application, require sequencing and monitoring. In-rush current precautions must still be taken as peripherals and their associated input capacitance are cycled on and off, inserted and removed.
Power management has evolved from one or two voltages controlled by MOSFET switches which are enabled by discrete logic circuits and ASIC controllers, to ASSPs like Hot Swap/Soft Start Controllers, Supply Sequencers and Trackers, Voltage Supervisors, Reset Generators and Watchdog Timers. However, comprehensive power-management design can become expensive and complex as different applications require different combinations and different versions of ASSPs.
Selecting the right combination of devices can be daunting, with hundreds of devices available from many different vendors. Understandably, designers often simplify their power management designs by ignoring certain possible fault scenarios, or by assuming that certain sequences will always occur.
One example is a power-management design that monitors only the input-supply voltage, and then implements the sequencing of other secondary voltages by tying the "power good" of one regulator to the enable of the next regulator. To be sure, this approach reduces cost and complexity by alleviating the need for a discrete sequencer as well as several precision voltage monitors to monitor each rail. Although this sequential approach reduces cost and complexity, power-supply failure response time can be significantly delayed, resulting in serious data corruption in the form of runt packets and the corruption of stored data.
PCIe voltages, currents and card input capacitance are defined for various slots. Table 1 shows the PCIe specification defining the +12V and +3.3V supplies and tolerances, capacitive loading and maximum current (including in-rush current) for different cards.
Table 1: PCIe power-supply requirements
PCIe also allows for hot-swap cards that require careful attention to limit the startup-voltage slew rate. Voltage supervisors should be used on the inputs to monitor the supplies to determine voltage slew-rate limiting. Although PCIe does not specify sequencing of power supplies, an individual application with secondary supplies can require complex sequencing.
Figure 1 shows the startup sequence of a PCIe card. A key specification shown by the arrow is the 100ms period which occurs after the card is inserted and the 12V and 3V power supplies are stable. After 100ms, the card is enabled by the PCIe bus host by releasing PERST# signal high.
Figure 1: PCIe startup waveforms
Often, the 100ms time is too short a period for the complete sequencing of secondary card supplies and the initialization of large FPGAs, ASICs and other configurable devices. Pulse stretching or delaying of the PERST# signal is often required to meet the individual requirements of each board.
Figure 2 shows a PCIe card power-down sequence. The PERST# initializes the shutdown, allowing devices to be powered down in a controlled manner before the power supplies decay.
Figure 2: Power-down waveforms
If cards are suddenly extracted while the socket is powered, devices will be powered down abruptly, which can cause catastrophic results. Care should be taken when designing boards so they can handle a surprise extraction and power down the board in a controlled fashion.
Numerous challenges must be addressed when designing PCIe power management. For example:
• In-rush current varies on each design and cannot, even momentarily, exceed the maximum PCIe supply-current specification. Both in-rush current magnitude and duration depend on board input capacitance and various other factors such as the startup currents of FPGAs or ASICs.
• Cards can require a unique hot swap controller circuit for each application.
• Timing may have to be extended beyond the 100ms PERST# signal to slow the reset timing, allowing for power-supply sequencing, FPGA configure time and CPU reset.
• The design should be fast enough to instantaneously respond and power-down the board during a hot-swap extraction without corrupting the system.
• All power supplies should be monitored for both under- and over-voltage conditions to maintain operating integrity.
• The sequencing of power supplies should be flexible, since it can be unique for every application and may need to be changed for design reiterations.
• Boards containing complex chips like CPUs usually require a stable core voltage before initializing the I/O voltages.
How are these challenges addressed? The traditional approach to designing power management for PCIe cards is to use a discrete solution. Figure 3 shows such an approach, where the hot-swap controller, sequencer, supervisors, reset generator and watchdog timers are all implemented separately.
Figure 3: Discrete implementation of power management
However, there are serious drawbacks to this approach. Discrete implementations require researching data sheets in order to choose from a wide selection of devices. Discrete designs are inflexible because any changes to the design, or a different application, will require a different set of discrete devices. Timing and control circuits that rely on R-C networks to establish their timing will change as components, age and voltage supplies vary. Finally, discrete designs are slow to respond to fault conditions such as surprise extraction, due to interoperability issues between devices from multiple vendors.
Integrating power management into a system would significantly reduce cost, not only by providing all power-management functions but also by eliminating duplicate functions, as those functions which share common resources could be combined.
For example, the multiple voltage supervisors, sequencers, hot-swap controllers, reset-generator ICs and trim-and-margin functions could be built into a single IC. A single, extremely accurate band-gap reference could be shared by multiple functions, further lowering cost without sacrificing accuracy or reliability.
More importantly, integration would eliminate communication time delays that occur with discrete solutions. Faults could be acted upon in tens of microseconds, instead of the hundreds of milliseconds typical for a microprocessor-monitored system. Trim, margin and voltage measurement could easily be accomplished by adding an ADC and a DAC.
ASICS are available that can combine some of the discrete devices required in power management. However, they typically require additional ICs, including a microprocessor, to complete the solution and will also include functions that are not required for the application. Further, the ASIC-based solution is hard to simulate and, being a “fixed” approach, it requires that any changes be made out of the board.
An alternative and more effective approach is to use a single, integrated power-management IC. By integrating all of the power-management functions, several key issues associated with discrete solutions are overcome. The intercommunication issues associated with separate devices from different vendors, and slow response to system fault conditions, are alleviated, as faults can be addressed in just a few microseconds. Overall cost is reduced as well, because critical functions are shared among several channels.
For example, the Lattice POWR1014A of Figure 4 integrates ten programmable voltage supervisors, all of which share a single, common band-gap reference, allowing all channels a 0.3% voltage-monitoring accuracy.
Figure 4: POWR10414A architecture
An internal clock and built-in digital timers resolve the inaccuracies associated with devices that use external R-C networks. Digital I/Os, programmable timers and a CPLD core allow the monitoring of PERST# and PRSNT# and generation of card-specific timing to ensure correct sequencing and configuration.
Additional signals can be generated, depending on the inputs that could alert the system to a reset or brownout conditions. The POWR1014A contains two charge pumps for controlling N-Channel MOSFETs. Hot swap can be easily customized for individual applications by varying the gate voltage and charge rate, while monitoring the system’s currents and voltages to maintain PCIe limits. The CPLD core allows easy modification of designs for various applications, board revisions and supplier variances. Inputs and outputs are easily configured and the CPLD core programmed through Lattice’s PAC-Designer design software.
PCI Express has standardized the interface and timing between the PC and add-on cards. The various applications require a customized design for each unique current, timing, voltage, and sequencing function. Discrete solutions are expensive, lack accurate timing, have low precision, have reliability issues due to their higher parts count and larger bill of materials, and are inflexible if the design changes. PMICs such as Lattice’s POWR1014A integrate PCIe power management into a precise, flexible, programmable, and low-cost solution.
About the author:
Jeff Hooker is a Senior Product Marketing Engineer at Lattice Semiconductor Corporation. Prior to joining Lattice, Jeff was a design engineer for Texas Instruments and has been in technical marketing and business development for Comlinear, National Semiconductor, Micrel and Microsemi. He holds a BSEE from the University of Washington and MBA from the University of Dallas. Jeff can be reached at email@example.com.
Article is courtesy Power Management Designline