Digital communication and control benefit power converter subsystems

Article By : Mark Hagen and Oleg Volfson

Understand how a power subsystem’s digital interface can benefit a specific electronic application.

Digital communication and control benefit power converter subsystems and the systems they are part of. A digital interface provides the power design engineer with greater flexibility to finalize and optimize certain parameters (e.g. sequencing of voltage rails, fault protection thresholds, and feedback compensation) after a system board returns from assembly.

Using the OSI model for digital communication, there are two major aspects of digital communication: the physical layer (PHY) over which communication is executed, and the data link layer, which defines the protocol or a command set that is used to communicate information.

The PHY for most power converters is the I²C serial interface, or its derivative, the SMBus, shown in Figure 1. The I²C interface defines a bidirectional clock signal (SCL) and a bidirectional data signal (SDA). The SMBus adds an optional alert signal (normally used to generate interrupts) and utilizes fixed logic levels for the signaling. Signal timing for clock and data are the same for both standards. These PHY choices allow multiple converter circuits to share a single interface with a host controller and are used in many server and PC subsystems, such as fan control and power/sleep functions.

The data link layer defines the information to be communicated between the host and voltage converter circuit. The data can be defined as a set of addressable registers with data bits uniquely defined for each IC, or the registers can follow an industry standard. One such standard for power management is the PMBus standard, which provides a defined set of registers/commands to communicate status. (For the distinctions between I²C, SMBus, and PMBus, refer to the PMBus website.)

Figure 1 The I²C/PMBus signals provide the physical interface most often used in communications to power converters.

The PMBus standard states that, to be considered compliant with the specification, a power device must implement at least one of the PMBus commands. Most PMBus-equipped devices implement the PMBus protocol for several common commands, such as setting the output voltage or reading die temperature. However, nearly all devices also implement MFG_SPECIFIC registers unique to that device.

Data and command formats

In addition, the PMBus standard calls out two methods to encode values into a digital word that can be communicated over the interface: direct and linear. For the direct method, the value in the command is the integer value the manufacturer has defined for the register in the device. Linear data format, on the other hand, is a form of floating-point value representation. In practice, most devices use either a Linear11 or Linear16 representation of data values.

The Linear11 data format has an 11-bit mantissa and a 5-bit exponent (Figure 2). Both the mantissa and the exponent are two’s-complement integers, meaning they can be positive or negative. You convert a real-world value to Linear11 format with Equation 1:

X_{REAL WORLD} = Y · 2^N (1)

Figure 2 The Linear11 data format has an 11-bit mantissa and 5-bit exponent, both expressed in two’s-complement.

The exponent N can be positive or negative, and 2^N defines the size of the Y mantissa’s LSB. Thus, the smallest number that can be represented using Linear11 format is ±2^-16 x 1 = ±15.3e-6. The largest number is ±33.5e6.

Since a signed, 11-bit integer goes from -1,024 to +1,023, the mantissa’s magnitude should be kept between 512 and 1,023 to maximize the data resolution during encoding. This suggests an approach to encoding a real-world value as a PMBus Linear11 formatted value: start with N = -16, then increment N until the mantissa’s magnitude is within the desired range.

The Linear16 format has a 16-bit mantissa and a 5-bit exponent, with both the mantissa and the exponent as two’s-complement integers. The data structure is like that of Linear11, except that Linear16 uses the full 16-bit I²C command packet for the mantissa and supplies the exponent separately (Figure 3), possibly mixed with a command code. For example, the MPQ4230 buck-boost converter from MPS uses the Linear16 format for VOUT_COMMAND and READ_VOUT, which set and read the output voltage, with the command code and exponent combined in one byte. You convert a real-world value to Linear16 format with Equation 2:

X_{real world} = Y · 2^N (2)

Figure 3 Linear16, like Linear11, uses two’s complement for the mantissa and exponent, but sends them in separate words that may also include a command code.

With a 5-bit exponent, the smallest number that can be represented using Linear16 format is still ±2^-16 x 1 = ±15.3e-6. However, the target value for the mantissa is now between 2¹⁴ and 2¹⁵, or between 16,384 and 32,767. This gives more resolution with which to set and read the output voltage.

PMBus commands can set the output voltage of a converter, enable the device, determine whether a warning threshold has been exceeded, and report a fault. Some commands are commonly used by many manufacturers, for example the STATUS_WORD command for MPS’s MPM3695-25 power module, shown in Table 1.

Table 1 This STATUS_WORD command provides an indication of various system faults.

But manufacturing-specific commands unique to a given device are also in use. Consider the MPM3695-25’s MFR_CTRL_COMP command. While STATUS_WORD is a common command for many ICs, the constant-on-time compensation loop for the MPM3695-25 is unique to this chip and has its own unique command, as shown in Table 2.

Table 2 This MRF_CTRL_COMP (Addr/Command 0xD0) command is unique to the MPS MPM3695-25.

Memory options for converters with a digital interface

Converters that can communicate over a digital bus also differ in terms of available memory options. These options fall into three memory classifications:

1. No permanent memory in the converter: In this case, the configuration is stored in the memory of the host microcontroller (μC), and the converter configuration must be loaded after every start-up. This can simplify the design and lower cost but requires a separate means to power the host μC at start-up so that it can configure the subject power converter. As an alternative, having the default registers in the digital converter sufficiently close to the final configured values may allow the converter to adequately power the μC while it finalizes the converter’s configuration. Consider, for instance, an application requiring fine adjustment of the core voltage for a μC core in a system-on-chip (SoC), or similar digital ASIC. A converter such as the MP8843 synchronous buck converter with an I²C interface must be reconfigured after each start-up, but defaults to a setting that will run the core during boot-up but not be optimized for processor speed and power dissipation. The application firmware in flash memory can, after booting, then write to the device’s I²C interface to optimize the core voltage as the system runs.
2. One-time-programmable (OTP) memory: OTP is a cost-effective, reliable solution for high-volume products needing to boot-up at an optimized setting. However, the approach may require the IC supplier to create a unique part number for each application. For example, the MPS MP8859 is a 4-switch buck-boost converter with an I²C interface and OTP memory that can set the output voltage, max current, switching frequency, and fault recovery behavior. A graphic user interface (GUI) program is available to select the register values that optimize performance in the user’s application. Once the complete set of register values has been determined, they can be permanently stored using the GUI to write to the OTP. These stored values then become the default start-up values for the registers, although the target system’s application software can still change the registers through the I²C interface after start-up. Once the designer has determined the desired default register values during end-product development, those values can be written into the converter OTP memory at manufacturing time, saving a step for the user
3. Non-volatile memory (NVM): NVM provides the highest flexibility, but at a higher cost. Users must also be aware of the type of memory used to implement NVM. Flash memory is widely used in μCs, but can degrade quickly when operating at high temperatures, which is often a concern for power converters. For this reason, companies such as MPS do not utilize flash memory in their products. The advantage of NVM is that register values can be saved to the NVM multiple times. This means that the hardware can be repeatedly prepped and tested as if it were being assembled/manufactured as new hardware. In these cases, users should store configuration information during product development exactly as they would during volume manufacturing.

Converter ICs with no permanent memory often do not follow PMBus standards for register definitions, which minimizes overhead for their configuration registers. With just a few optimized registers in such devices, the user’s effort to develop firmware to read and write to these non-standard registers is manageable. However, developing the firmware to interface to a device with dozens of unique configuration and monitoring registers can be daunting, so making the register definition follow PMBus standards can be very valuable.

Adding a digital control and communication feature to the power converter allows better integration of these devices into modern electronic systems by streamlining system design, and increasing flexibility and reliability. However, the best way to reap the benefits of these features is for a design engineer to familiarize themselves with different approaches to converter designs with the digital interface and control, as well as the specific nomenclature of these devices. Understanding the communications standards and the tradeoffs associated with following them will help designers choose the right combination of cost and flexibility in a power converter.

Mark Hagen is a Senior Field Applications Engineer for Monolithic Power Systems.

Oleg Volfson is a Senior Field Applications Engineer for Monolithic Power Systems.

Related articles:

• Power supplies go digital
• PMBus-compatible PoL regulator for active power management
• System configurations for power systems based on PMBus 1.3
• Time to refresh PMBus
• Power line communication implementation for DC applications

Subscribe to Newsletter