1

I'm designing a router and have noticed that some designs use a 'MPU with built in PHY' package on a single chip, whereas others use a PHY IC communicating to a separate CPU/MPU IC over PCIe.

What's the point in separating these two devices? I would think that having two in one package is the best way to go.

ocrdu
  • 8,705
  • 21
  • 30
  • 42
cdubs
  • 413
  • 2
  • 9
  • 1
    If the device connects over PCIe its more than just a PHY - its the whole shebang including the mac. – Kartman Sep 28 '22 at 05:31
  • 1
    No one mentioned so far: PHY may heat considerably, and internal PHY will also heat the other parts of the chip. – Anonymous Sep 28 '22 at 12:00

2 Answers2

1

The PHY that is typically implemented within an MPU is not as sophisticated or flexible as an external one.

For example the PHY in an MPU may only support 100BaseT and not 100Baset1, 1000BaseT1 or 10BaseT1-S. In such cases you may be able to use the MAC internal to the MPU then the standard MAC to PHY interface such as RGMII or SGMII to connect to the PHY.

Also other features such as 802.1AS (Precision Time Protocol) may not be supported.

Kevin White
  • 32,097
  • 1
  • 47
  • 74
1

What's the point in separating these two devices?

Benefits of joined-up

  1. Smaller footprint
  2. Known compatibility
  3. Manufacturer support

Benefits of separate

  1. Can use ubiquitous parts
  2. More examples
  3. Better supply availabilty

Importance for long-term

This last issue (better supply possibilities) is exceptionally important in designing long-term devices, where you want the maximum amount of compatibility with what will be available in the future, ideally with manufacturer support. You can still get Z80 chips, for example, which will plug straight in. If you picked something perfect-but-specialised, you'll be out of luck, and will have to do a redesign, will may cascade into all kinds of difficulties.

Of course, in times like the present (2022) there are widespread chip shortages and many exceedingly common chips are unavailable (ATMega CPU, for example, is unavailable from the major distributors I just checked, which normally have 10,000 or more units; same for ENC28J60 ether.) It has to be said though: you are still more likely to find these eventually compared to a part which didn't sell well.

As an example, one of my clients is currently in a crisis because a common part is unavailable, which led to the subsystem manufacturer cancelling a product line, which has triggered a casing crisis because the alternatives don't fit; it's non-trivial because these are sealed underwater units. My client is nowhere near as large as those in the car industry where they can't get chips either: that industry is large, but low-margin and inflexible (because of certification).

jonathanjo
  • 12,049
  • 3
  • 27
  • 60