Which Ether PHY to use..

[mrpackethead]

I'm working on a new design project and have to choose which etherphy i'm going to use..  I'm wanting the most robust PHY i can use, as its going to get used in horrid conditions..   

Options i'm looking at are LAN8701a ( Microchip ) or DP83848C ( TI ).

The TI chipset has worked for me, although in POE designs i am still getting failures.. I suspect this is due to common mode spike on the PHY side of the device.   I've got TVS diodes on the LAN side of the transformer and that stops diferential faults.. I'm redesgining to put both in.

 

[MagicSmoker]

I'm working on a new design project and have to choose which etherphy i'm going to use..  I'm wanting the most robust PHY i can use, as its going to get used in horrid conditions..   

Options i'm looking at are LAN8701a ( Microchip ) or DP83848C ( TI ).

The TI chipset has worked for me, although in POE designs i am still getting failures.. I suspect this is due to common mode spike on the PHY side of the device.   I've got TVS diodes on the LAN side of the transformer and that stops diferential faults.. I'm redesgining to put both in.

I don't have any experience with PoE, though I am considering it for a future motor drive application. I do use the LAN8710A PHY in a high-noise environment with excellent results, but only after a design revision where I paid more attention to the rules of ethernet PHY layout: minimize the distance of the differential pairs between PHY and mag-jack; ensure that the differential pairs are routed as such; place the termination resistors for the differential pairs as close to the PHY as possible without violating the other two rules.

I imagine you will have a lot more problems with signal integrity if you try to do PoE on the signal pairs, instead of the unused pairs, and try to protect against overvoltage on those pairs with TVS diodes or the like.

[sarepairman2]

diethyl ether

[nctnico]

I'd choose TI over Microchip any day! Regarding the spikes: it is good to use TVS diodes on the signal lines in PoE applications. I'd put them on the chip side of the transformer though.

[mrpackethead]

I'd choose TI over Microchip any day! Regarding the spikes: it is good to use TVS diodes on the signal lines in PoE applications. I'd put them on the chip side of the transformer though.

Why the preference for the TI product...

Anything else you'd do to protect the Phy?

[AndyC_772]

The DP83848 is a very old National part. There's nothing 'wrong' with it as such, but there are much smaller, lower power and probably cheaper parts available elsewhere.

Try the KSZ80x1 range from Micrel.

[mrpackethead]

The DP83848 is a very old National part. There's nothing 'wrong' with it as such, but there are much smaller, lower power and probably cheaper parts available elsewhere.

Try the KSZ80x1 range from Micrel.

Yes, that part is really old. .  Interesting Microchip recently aquired Micrel.   The Part i've been playing with is the LAN8740, its considerably smaller..  But is it as robust?

I attached a schematic of what i'm proposing for the TVS diodes...  both sides of the transformer.. The schematic symbols really are bad though i shoudl update them!

[nctnico]

I'd choose TI over Microchip any day! Regarding the spikes: it is good to use TVS diodes on the signal lines in PoE applications. I'd put them on the chip side of the transformer though.

Why the preference for the TI product...

Microchip competes on price and not on quality. If your product needs quality then Microchip parts have no place in it.

[mrpackethead]

I'd choose TI over Microchip any day! Regarding the spikes: it is good to use TVS diodes on the signal lines in PoE applications. I'd put them on the chip side of the transformer though.

Why the preference for the TI product...

Microchip competes on price and not on quality. If your product needs quality then Microchip parts have no place in it.

Thats a pretty broad brush statement. I've used 10's of thousands of PIC32 micros, lots of linear regulators and lots of other bits and peices, and they have proven to be reliable and met the requirements just fine.    If truth is told, i've probably had more problems with TI parts than Microchip parts.. 

what particular attributes cause you to say this?

[JoeN]

Try the KSZ80x1 range from Micrel.

Ahem...   Microchip!

[mrpackethead]

Try the KSZ80x1 range from Micrel.

Ahem...   Microchip!

yes, they do seem to be buying up a lot of stuff.    Micrel have some pretty handy small switch chips at very resonable prices.

[nctnico]

what particular attributes cause you to say this?

Compare specs for similar parts and push parts to the limit. If your applications can live with cheaper, less well designed parts then Microchip will work OK. If you push parts to the limits (while staying within spec ofcourse!) you'll see more failures & quirks with the cheaper B brands like Microchip.

[Siwastaja]

Compare specs for similar parts and push parts to the limit. If your applications can live with cheaper, less well designed parts then Microchip will work OK. If you push parts to the limits (while staying within spec ofcourse!) you'll see more failures & quirks with the cheaper B brands like Microchip.

Any specific examples? Maybe this could be another thread, but I'm interested why I shouldn't choose Microchip parts - they do have many really useful products. Of course the products must work when used within limits.

[D3f1ant]

Microchip bought Micrel this year ($800m), and most if not all of Microchips ethernet and usb products are a result of their  SMSC acquisition, which they bought in 2012 (for nearly $1b).
Nothing wrong with the quality of any of these products, or any microchip product I've ever used.

[JoeN]

Microchip bought Micrel this year ($800m), and most if not all of Microchips ethernet and usb products are a result of their  SMSC acquisition, which they bought in 2012 (for nearly $1b).
Nothing wrong with the quality of any of these products, or any microchip product I've ever used.

All of these products are available on their samples page.  If you want to try before you buy they are quite liberal.  They will send you 3 of two parts per order and you can order a couple of times a month if necessary.

[Siwastaja]

I have just some totally horrible experience from an SMSC product. It was an integrated MAC+PHY. The product was basically unusable, I worked trying to get it work for months and never got it running reliably. It was a huge waste of time. And I was just porting something I had done earlier on DM9000A, which was order or two orders of magnitude easier.

The SMSC part was complex as hell, with documentation worse than many Chinese datasheets.

Emulating an I2C interface by writing "clock" and "data" bits to a configuration register, only to access more configuration registers, is something so poorly designed that I wanted to cry. I understand why they did it, but having a huge burden of history on your shoulders because you only glued some random old designs together quickly to release a "new" product, means you cannot call it "integrated product" or well-engineered.

They may have better products in their portfolio, and maybe this was a single failure, but I'll never want to use LANxxxx anything. If nothing else, I still get shivers from the name alone.

[mrpackethead]

Which product was this?

I have just some totally horrible experience from an SMSC product. It was an integrated MAC+PHY. The product was basically unusable, I worked trying to get it work for months and never got it running reliably. It was a huge waste of time. And I was just porting something I had done earlier on DM9000A, which was order or two orders of magnitude easier.

The SMSC part was complex as hell, with documentation worse than many Chinese datasheets.

Emulating an I2C interface by writing "clock" and "data" bits to a configuration register, only to access more configuration registers, is something so poorly designed that I wanted to cry. I understand why they did it, but having a huge burden of history on your shoulders because you only glued some random old designs together quickly to release a "new" product, means you cannot call it "integrated product" or well-engineered.

They may have better products in their portfolio, and maybe this was a single failure, but I'll never want to use LANxxxx anything. If nothing else, I still get shivers from the name alone.

[nctnico]

They may have better products in their portfolio, and maybe this was a single failure, but I'll never want to use LANxxxx anything. If nothing else, I still get shivers from the name alone.

Their USB based ethernet chips are OK though as long as you have the resources to do a respin or two to solve the EMC issues.

[Siwastaja]

Which product was this?

Can't remember. It's the one in the older (maybe 2004-2005) super expensive Altera Statix II FPGA "DSP" development board.

I'm sure it works if you use it with a CPU and use the drivers that are written and debugged once to work, then reused over and over again. I had to do it on custom HW only, and it was a nightmare.

[mrpackethead]

Yeah, its easy to use with a small uP.

[Scrts]

I've used Micrel on industrial design (broadcast television equipment) and they didn't fail any time. We've deployed 10/100/1000 devices as well as 10/100 ones.
Once you have hardware correct, the PHYs are working fine. I've also made 10/100/1000 board passing EMC on the first time, however you have to bury as many wires as you can and don't use 25MHz XTAL as the internal generator will radiate a lot. Supply the clock from somewhere else. In my case - FPGA was the source.