When should you use a Point of Load Converter (POL) with PMBus, and when should you use a Digital Manager with old school POLs? In this post I will provide some heuristics for making this choice.
So as to not mislead, my goal is modest: to provide enough intuition to make a first pass choice, that becomes an area to explore in more detail. This is no different than rethinking where you have been for the last 24 hours before looking for your lost car keys, because you can’t look everywhere at once, and you want to improve your chances of finding them.
First let’s draw some diagrams of these devices so we can examine their basic differences. Sometimes pictures make things clearer.
POL with PMBus
On a datasheet a POL with PMBus might be more like: Dual Output … Step-Down DC/DC Controller with Digital Power System Management. Some of the structure of the devices ends up in the title, but the basic overall form is reasonably consistent.
If we draw an architectural line right down the middle of the device, the left side has all the logic for the PMBus, Telemetry, Supervisors, and the DAC driving the reference input to the POL. On the right side is the Point of Load Converter (POL). Even though we split the functions, they are both in one package and purchased as one device.
Digital Manager with POLs
A Digital Manager does not have a power conversion stage and relies on external POLs. It is a multi-package solution.
The manager controls each POLs output voltage, current limit, measures outputs and input, and handles faults. Essentially, the POL with PMBus model was cut at the dashed line. Typical managers handle 4 or 8 POLs.
Implications of splitting the POL with PMBus
The obvious implication of splitting the functionality (which is not really something new because POLs existed long before the managers), is independence of functions. The POL market is quite large and diverse and supports a very broad range of solutions. We often think of POLs as switching converters (SMPS), such as a Buck, but LDOs are just as controllable and observable. Therefore separation of the functions creates a very large number of design combinations.
On the other hand, because package costs are significant, a POL with PMBus may have a lower cost per channel. By not requiring pins between the management functions and POLs more lines of control allows a higher level of functionality and integration. Performance will also exceed the performance of a divided solution in some dimensions.
This is a classic case of performance vs. flexibility. However, for a large multi-rail solution, a global optimum may not follow this simple tradeoff. So let’s explore further…
My advice is always look at the POLs with PMBus first. These POLs tend to address particular markets and if you fit into one of these niches, you can benefit from the tight integration, simplicity, and performance. If your needs don’t fit, then look at Digital Managers.
Let’s take a couple of examples.
Suppose our design needs the following rails:
1V / 20A
3.3V / 8A
5V / 5A
1V / 40A
5V / 100mA
3.3V / 10A
3.3V / 200mA (low noise)
In this example there are large differences in power level, and one of the 3.3V rails has to be low noise. The low noise 3.3V rail warrants an LDO. Suppose the 5V rail does not fit well into the POL with PMBus. You can do it, but you may pay too much for the device, or the board space may be too large. Maybe for the 1V and 3.3V higher power rails you have trusted existing solutions. A Digital Manager like the LTC2974 and 4 POLs would be a good solution.
What about very high rail count systems? If there are a lot of rails, chances are some rails should be handled like Example 1, and some like Example 2. The good news is both designs use the same PMBus, so from a firmware perspective, the world just looks like one PMBus with lots of rails.
Making Things Play Well
In my sequencing blog post I talked about handling time synchronization and faults. In order for these different design structures to play well, it is helpful to use devices that have a consistent approach to configuration, sequencing, fault management, and current sharing. Not all vendors use a consistent set of mechanisms, so the following is a list of things you should look for.
These items will help:
- A shared clock so all devices run at the same internal clock rate
- A synchronization mechanism so all devices have the same time zero
- A shared fault mechanism so all parts know when a fault occurs
- A shared start/run/reset mechanism so all parts start together
- A consistent approach to current sharing
- A shared alert pin so that the PMBus host knows when any rail faults
- A shared PMBus so the PMBus host can control all rails
- A single configuration tool so that all rails are managed in one display
The PMBus Specification does not address the top 5 items in the list, or the last item, and de facto standards don’t exist. This means you need to be a bit careful when mixing and matching devices from different vendors. In fact, many times you can’t mix and match within one vendor, so you need to check this list against any product portfolio up front before committing too many resources.
Note: there is one caveat: if you are using one vendor’s POLs with PMBus, and the same vendors Digital Manager, it is quite easy to integrate POLs from any vendor. The Digital Manager allows you to glue in just about anything with a feedback pin. So you are never trapped. For this reason, in the long run it is helpful to start with a vendor that has a good family of Digital Managers if you use their POLs with PMBus.
I presented two generic design choices, a POL with PMBus added on, and a Digital Manager that can take care of multiple POLs. These basic designs can also be mixed together in larger systems. Tradeoffs involve natural divisions in the functions, such that POLs with PMBus are optimized for niches and can be a very high performing, but Digital Managers offer a lot of flexibility.
This flexibility is not necessarily higher or lower cost, because the total cost depends on which POLs are used with them. Systems built with either design or a mixed design benefit from a consistent strategy for synchronization and faults that are not covered by the PMBus itself. Having a good Digital Manager allows mixing in parts from different vendors.
The bottom line is:
- POLs with PMBus, Digital Managers, and traditional POLs, are the system building blocks
- Consistent mechanisms make integration much easier, especially sequencing and fault logic
- Good Digital Managers allow integration of any POL including LDOs