In this series of blogs so far, we have discussed the internal extremes of industrial power levels: from kW motor drives to µW energy harvesting for IIoT node power. The industrial environment can certainly be varied, with hundreds if not thousands of amps flowing around power lines, and voltages up to 690VAC inside buildings. Surges and transients are routine as loads switch and contactors operate, radiated and conducted EMI is at a high level, and temperatures and vibrational forces can be intense. In the middle of this maelstrom of environmental effects, CPUs, FPGAs and other forms of data processing circuitry keep control of the whole operation with expectedly high reliability, maintaining productivity and minimising downtime.
To achieve this, the local environment of supply rail stability, electrical noise and temperature must be precisely controlled, even though the ICs themselves are not helping the problem, taking tens of amps, sometimes off sub-1V rails with large load steps.
When process control involved switching motors on and off, and digital logic was TTL powered from 5V, it was feasible to have a centralised power architecture (CPA). A single cabinet AC-DC converter supplied the low voltage power rails for cards in a control cabinet rack. Load currents were low, voltage drops manageable, and the noise margin of the logic tolerated the pick-up on the long leads. However, as processor speeds and power draw increased, voltage rails dropped to 3.5V then 3.3V. The CPA scheme became unworkable and a new distributed power architecture (DPA), was adopted.
Here, a higher voltage, usually 24V, is routed around a cabinet, and board-mounted DC-DC converters step the voltage down to the end-load requirements.
The converters, typically isolated, ensure high ground loop currents are kept local to the load, thus minimising interference. Although expensive, there were advantages that a 24V supply could feature battery backup. Cards, duplicated in a redundant configuration, could allow ‘hot swapping’ if any failure occurred.
As processing demand increased, the need for multiple rails increased, and core voltages headed towards 0.6V. More supply rails meant more isolated DC-DC converters on each board.
Intermediate bus architecture (IBA) provides a solution to this challenge. Widely used in data centres, IBA uses a single, isolated converter to generate a bus voltage for each card (perhaps 12V) and then low cost, non-isolated point-of-load (PoL) converters provide the final voltage. The intermediate bus converter (IBC) can be powered from the traditional industrial 24V rail, though often now this is much higher (48V or even 380V) to reduce current and resistive losses. The IBC can also be unregulated, saving system cost, as PoL converters are generally wide-input.
PoL converters provide an oasis of (electrical) calm for the end load
So, what do we now require of PoL converters? They typically must provide an output which stays within ±3% of nominal (including initial tolerance, noise spikes and voltage steps due to load transients). They must be extremely compact, as they are physically close to the typical load of a CPU or FPGA with their myriad of address and I/O connections. They must also be efficient, so heat losses do not add significantly to the load power dissipation. There will be at least two voltages needed, core and I/O, with usually several more included in a typical circuit. Figure 1 shows a ‘power tree’ for a Xilinx system-on-chip (SoC) as an example. As shown, it can be convenient to cascade PoLs using the output of one to feed another.
Figure 1: A power tree diagram for a Xilinx Zynq7xxx SoC
PoLs can be implemented as a control IC on the mainboard with additional discrete components, typically an inductor with input/output capacitors. The MAXIM MAX17760 is a good example, with its 4.5V to 76V input and programmable output down to 0.8V at 300mA. It comes in a tiny 3x3mm TDFN package. For higher currents, one could consider the VICOR PI332x-00 control IC. With versions available down to 3.3V, and a 22A output. A low-drop linear regulator such as the STMicroelectronics ST730 could be employed to achieve the final stage of regulation, presenting an efficient and cost-effective solution where the input-output differential voltage is low. For example, generating 2.1V from 2.5V with this LDO results in an 84% conversion efficiency.
Additional benefits of PoL
When circuit board real estate is at a premium, and minimum inventory and placement costs are essential, a PoL module is a right solution. Often packaged in a SIP format, they save space and have the advantage of a pre-tested, fit-and-forget high-performance converter. An example is the XP Power TR20 series rated at 2A with outputs available up to 15VDC and with an efficiency of 96 percent.
Providing a quiet electrical environment for today’s complex ICs is best achieved with PoL converters. Designed for the specific application, they will often have the additional features of output sequencing and dynamic programmability. Allowing output voltage adjustment, or even control loop response for maximum flexibility and processor performance.
High-end products also incorporate digital interfaces, typically I2C with PMBus commands for programming of monitoring threshold levels and performance.
With the massive increases in data processing seen in the industrial environment, the need for clean, conditioned power has never been greater. Intermediate Bus Architecture (IBA) and Point of Load converters facilitate these requirements elegantly and cost-effectively. IBAs provides an isolated single-ended source of DC power. Cheaper non-isolated DC-DC PoL converters then create multiple rails, in some cases also adding output sequencing and dynamic programmability as a feature. Available in different packages, they can save board space, and offer a fit and forget DC power solution.
In the final blog of this series, we will look at what power conversion technology might bring to industry over the years ahead.
Other articles in this series