Multiple power rails design
robertjzima , 02-25-2017, 02:43 PM
Hello. I've just started a "small" project 
. The point is to design s baseboard with Xilinx FPGA Spartan-6. The FPGA will be connected to iMX6 Rex using PCie. UART, etc. It's like imx6 baseboard with FPGA on it. I've never done that complex project. It's quite complicated and I need your help, if it would be possible . I thinks it's going well, but I've just stuck in power supply design. The problem is that I've never done multiple rail design. I know how to connect regulators and compute In/Out capacitance, inductance etc, etc. I'm just not sure, if multiple design rails have some rules or recommendations on how it should be done. I read a spartan datasheet very thoroughly. There is no need for power sequencing and high speed transceivers must have some filters, nothing special. The power tree is in the attachment
There are 4 different FPGA I/O banks Vccio #1-4, Vccint is FPGA internal supply, 3V3_RAIL and 5V_RAIL is rail for powering other parts on the board which are not related to FPGA.
What do you think about it? Thanks
. The point is to design s baseboard with Xilinx FPGA Spartan-6. The FPGA will be connected to iMX6 Rex using PCie. UART, etc. It's like imx6 baseboard with FPGA on it. I've never done that complex project. It's quite complicated and I need your help, if it would be possible . I thinks it's going well, but I've just stuck in power supply design. The problem is that I've never done multiple rail design. I know how to connect regulators and compute In/Out capacitance, inductance etc, etc. I'm just not sure, if multiple design rails have some rules or recommendations on how it should be done. I read a spartan datasheet very thoroughly. There is no need for power sequencing and high speed transceivers must have some filters, nothing special. The power tree is in the attachment There are 4 different FPGA I/O banks Vccio #1-4, Vccint is FPGA internal supply, 3V3_RAIL and 5V_RAIL is rail for powering other parts on the board which are not related to FPGA.
What do you think about it? Thanks
mairomaster , 02-27-2017, 02:13 AM
In my opinion it is a bit risky to start such a project if you don't have any experience with multi-rail design, unless you have a good supervision. Having 10 different supplies for the FPGA only makes it easy to make a mistake somewhere.
I didn't notice anything really wrong with the power tree, but I didn't have much time to look at it.
I didn't notice anything really wrong with the power tree, but I didn't have much time to look at it.
robertferanec , 02-27-2017, 11:10 AM
Robert, I would recommend to follow the reference design (especially if you feel, that you do not have enough experience). Are you trying to design your own power management or this is from the reference schematic?
robertjzima , 02-28-2017, 08:15 AM
It's something between reference design from Xilinx and Reference schematics from Digilent. Digilent makes a lot boards based on Xilinx FPGAs. Their schematics are public. Reference XILINX design is quite old. Lot of ICs, regulators, etc are even EOL or not recommended for new design. I can't just use all of these chips. Another problem is that I need to power more things, not just FPGA and its peripherals. These two pictures in the attachment are autogenerated by TexasInstruments Webech tool. There you can set your concrete FPGA, like Spartan 6 45T, and all your power requirements. Webech then generate a block diagram, like those two pictures in my first post. It also generates complete and working schematics based on TI Regulators. It's very popular tool.
Well, I got probably only one question. Is it better to have "core" rails (in this FPGA it's: Vccint = 1V2, Vccaux = 2V5, Vcc02 = 3V3) regulated by one or separate controllers? Documentation doesn't mention these things but I think it's important. In the reference schematic they use only one Power controller (pic in attachment) for all of them. As you can see, they have a power controllers which are quite expensive and nearly EOL. Lot of newer designs, like those from Digilent, just use separate power regulators for every rail. But some of them have, for example two 3V3 regulators. One for Vcc02 rail and another one for other 3V3 things, peripherals, etc. Which approach is better?
Well, I got probably only one question. Is it better to have "core" rails (in this FPGA it's: Vccint = 1V2, Vccaux = 2V5, Vcc02 = 3V3) regulated by one or separate controllers? Documentation doesn't mention these things but I think it's important. In the reference schematic they use only one Power controller (pic in attachment) for all of them. As you can see, they have a power controllers which are quite expensive and nearly EOL. Lot of newer designs, like those from Digilent, just use separate power regulators for every rail. But some of them have, for example two 3V3 regulators. One for Vcc02 rail and another one for other 3V3 things, peripherals, etc. Which approach is better?
robertferanec , 02-28-2017, 08:40 AM
I have used ISL6236A regulator a lot. It is not new, but it is very flexible, you can set up or tweak output voltage, it can deliver high currents, it has all the control signals, maximum current limit, it is small, it can have wide input range or run from 5V. So, you can use one type of regulator to create maybe even all your voltages.
I would not use different chip for each power and I would really consider what are the required maximum currents. From the block diagram it looks to me like some of the regulators may be designed unnecessary big (3.3, 13A, 5V 13 A, ... Input voltage 24V/5A ??? ). Personally, when I need to change power supplies I follow reference schematic, check the voltage and current and replace it.
Using one regulator or multiple regulators is up to you. Some manufacturers design a one multi output dedicated regulator for particular chip. It is a nice solution with the right maximum currents, right number of outputs and usually it is smaller than individual regulators. But you need to be sure you will be able to buy it for couple of years - which is not always the case. On the other hand, if you go for individual regulators, you can use something generic - it usually requires more space, it may be more expensive, but it is more flexible and usually the chips are manufactured for longer time than the special chips.
I would not use different chip for each power and I would really consider what are the required maximum currents. From the block diagram it looks to me like some of the regulators may be designed unnecessary big (3.3, 13A, 5V 13 A, ... Input voltage 24V/5A ??? ). Personally, when I need to change power supplies I follow reference schematic, check the voltage and current and replace it.
Using one regulator or multiple regulators is up to you. Some manufacturers design a one multi output dedicated regulator for particular chip. It is a nice solution with the right maximum currents, right number of outputs and usually it is smaller than individual regulators. But you need to be sure you will be able to buy it for couple of years - which is not always the case. On the other hand, if you go for individual regulators, you can use something generic - it usually requires more space, it may be more expensive, but it is more flexible and usually the chips are manufactured for longer time than the special chips.
robertjzima , 02-28-2017, 10:48 PM
Thanks! Very helpful answer. The picture in my previous post was from official Xilinx reference schematic. They use even more output current, 20A at 3V3, 20A at 1V5 and all "core" rails are set to 10A. It's a mystery 
. Btw, 24V 5A is not very accurate. The Webench system set this automatically when I've chosen input voltage range from 7V to 24V.
. Btw, 24V 5A is not very accurate. The Webench system set this automatically when I've chosen input voltage range from 7V to 24V.Use our interactive Discord forum to reply or ask new questions.