If you need a spare RJ45 switch port but the built-in ones are all in use, don’t sweat it if you’ve got a spare SFP instead. Ethernet adapters can’t be beat, but (speaking of sweat) mind the heat.
My mid-September initial tutorial on small form-factor pluggable (SFP) modules covered their traditional interfaces to fiber optic cables, with subsequent single- and multimode module teardowns following in short order. It also looked at adapters enabling interface between a SFP cage and conventional Ethernet cabling.
Here’s an example adapter image from that post.
And here’s the relevant associated prose, with bolded emphasis for today’s thematic purposes:
RJ-45 modules mate the SFP or SFP+ circuitry to an Ethernet transceiver. Speeds up to 10 GbE, such as with the module shown at the top of this section, are widely available. These modules tend to run “hotter” than fiber optical or basic wire alternatives, all other factors being equal.
I also touched on the incremental heat dissipation of a module containing a SerDes-interfacing Ethernet transceiver versus one that more simply mates to an optical transmitter-plus-receiver combo (apparently in spite of the fact that the latter includes a laser) within a subsequent blog post that discussed the migration to multi-Gbit Ethernet equipment.
I’ve also seen some online commentary suggestive of high operating temperatures and associated questionable long-term reliability, particularly of the devices’ SFP+ subsystems [and double-particularly in conjunction with the use of SFP+ Ethernet adapters, I’ll add in arrears]. To at least some degree, this discussion is likely reflective of the switches’ fan-less designs.
And just last weekend I stumbled across another case study example of the high-temperature trend. I was reading up on reviews of USB- and Thunderbolt-to-Ethernet adapters, specifically various units from QNAP, both for further conceptual research purposes and in preparation for my own LAN migration plans. QNAP’s latest-generation adapters mate USB4 and Thunderbolt 4 to either 10G SFP+ or 10 GbE (and in both single-and dual-port forms); the prior-gen adapters I happened to be perusing at the time, the QNA-T310G1S (10G SFP+) and QNA-T310G1T (10 GbE), were Thunderbolt 3-based and single-port only.
Here are stock shots of both in sequential order.
Specifically, the reviews I was reading were from ServeTheHome and NASCompares; both conveniently included teardowns. But as I scrolled through them, I was initially confused. The text and pictures in the ServeTheHome piece both trumpeted a device that notably relied solely on only a passive heatsink, while those in NASCompares’ writeup highlighted a device containing a more sizeable heat plate accompanied by an attached active fan.
Match up the image to the associated color scheme (not to mention external port), however, and the seeming discrepancy will clear up in short order. ServeTheHome disassembled the 10G SFP+ version of the product family, whereas NASCompares went with the 10 GbE adapter.
Spec divergence
With the QNAP devices as a representative case study, I’d therefore expect to see some further thermal dissipation assistance within today’s Ethernet-augmented dissection patient as compared to its optical-only predecessors—no room for a dedicated heatsink, presumably, nor even more certainly for a fan, but presumably at least some thermal paste and/or padding. But a further surprise awaited me even before I got inside.
Here’s the stock shot of what I thought I’d bought, as usual from Amazon’s Resale (formerly Warehouse) used-goods section: an Ipolex ASF-GE-T 10/100/1000BASE-T copper SFP to RJ45 transceiver with auto-negotiation support.
What Amazon shipped me instead, I learned, was the Ipolex ASF-2G-T. Although I didn’t realize this until it was in front of me, still in beat-up packaging, ready for teardown (not to mention beyond the retailer’s 30-day return “window”). As usual, it’s accompanied in the following photos (as well as other to follow) by a 0.75″ (19.1 mm) diameter U.S. penny for size comparison purposes.
My immediate deduction, as I’d unfortunately encountered no shortage of times before, was that someone had purchased a new Ipolex ASF-GE-T, swapped it for an ASF-2G-T already in their possession (still functional or not, not to mention authentic or not…who knows?) and returned the doppelgänger for full refund, with Amazon none the wiser as to the subterfuge it’d been the victim of…not to mention me, who’d then bought the stand-in. That said, while I likely still could have gotten my money back if I complained strongly enough, I didn’t much care in this instance. It had only cost me $8.50, and I was going to destructively (presumably) disassemble it anyway.
But bigger picture, why would someone even exchange a 2.5GbE ASF-2G-T for a 1 GbE ASF-GE-T? At first glance, the stand-in (assuming it’s still working, that is) is the seemingly superior option, at least from a bandwidth standpoint. My working theory neatly exemplifies a broader caution I’d already planned on passing along regarding adapters like these.
The Ipolex ASF-2G-T is no longer even listed on the supplier’s website, or anywhere elsewhere on the Internet (save for, maybe, the Internet Archive), for the matter—at least I wasn’t able to find it (let me know if you have better luck). But a more general Google search on the term “ASF-2G-T” reveals it’s still available for purchase from 10Gtek, an Iponex competitor.
Look closely at the specs and you’ll see that “2.5G” is the only data rate listed for 10Gtek’s version (and presumably also the Ipolex version) of the ASF-2G-T. In contrast, a different variant, the ASF-2G2-T, supports both 1G and 2.5G data rates. And the ASF-GE-T I thought I’d bought? It’s, again, a multi-rate “10/100/1000BASE-T copper SFP to RJ45 transceiver with auto-negotiation support”.
When evaluating SFP Ethernet adapters, keep in mind that they don’t necessarily support multiple possible data rates, for compatibility with whatever’s connected to the other end of the Ethernet cable, far from offering auto-negotiation handshaking capabilities (versus manual configuration, which necessitates the use of a managed switch).
Retention implementation deviation
Before trying to figure out how to get underneath the Ethernet adapter’s metal enclosure, I’ll first share the rest of the exterior shots.
Right side, as before with “right” referencing its orientation when inserted in a network switch or other gear’s SFP port.
Bottom. As before, note the retaining pin, although this time it’s not retractable.
Instead, in this case, rotating the “lock” mechanism lever in the front of the module moves a “ramp” into position adjacent to the pin, enabling you to then slide the module out of the cage.
Left side.
Back to the RJ45 connector for the more traditional right-side-up and upside-down views.
And the same goes for the SFP connector on the other end.
Heavy metal wrestling
Now to get inside. First step; peel away the topside sticker and see if there’s a screw head underneath. Nope.
Past modules had small inward-bent pins on the sides that, when coaxed back out of the enclosure, enabled the two halves to then slide apart. This time, the “extra” metal pieces were more sizeable and looked more like cage-retention “springs”, but I tried anyway. Again, nope.
Finally, giving the assemblage one final frustrated look-over, I belated noticed two small retention pins alongside the SFP connector.
Like the rest of the case, they were made of sturdy stuff, but I eventually “sprung” them both.
As prognosticated beforehand, there’s the thermal pad. The first of several? Hold that thought.
It originally covered both ICs on the right half of the PCB topside, as the slight indent matching the shape of IC U7 reveals. The plastic shell surrounding the SMD-packaged LAN transformer on the PCB’s left side got mangled by yours truly while wrenching the topside metal piece away.
Now how to get the PCB the rest of the way out? First remove the lock mechanism, mebbe?
Bingo (top view of the PCB extraction process first, then bottom).
Leaving behind yet another, smaller thermal pad.
PCB insights
Let’s look again at the comparatively boring RJ45 and SFP ends first, first right side up and then upside down in both cases.
Now for the perspectives you all really care about. This (top) side you’ve previously sorta-seen already, in encumbered-by-assembly form.
The LAN transformer is the AE-SE24203 from Apps Electronics. To its right (I hope you’ll take my word for it given the faint markings) is the primary thermal pad rationalization, Motorcomm’s YT8821C single-port, 2.5 GbE physical layer (PHY) transceiver. And to its right, past the inductor intermediary, is the aforementioned U7, in a diminutive five-lead package (which always leads me to suspect a dual-transistor structure inside) and faintly mixed-case labeled as follows (readers?):
JWQZY
4y6eE
Now for the other side (both flipped and rotated; note that the RJ45 is now on the right).
Passives dominate the landscape this time. At far left is a three-lead (single transistor?) packaged device labeled as follows (with the last digit 90° rotated versus the others).
P5S4
To its right is the much larger IC originally with the smaller thermal pad atop it and labeled as follows (first three lines):
F330GB
F107079
PJ2423
Those three lines’ worth of characters weren’t in and of themselves terribly helpful from a Google search standpoint, but the last line was more enlightening (at least in theory):
GD ARM
My suspicion is that the first two characters reference the vendor GigaDevice, with the other three indicating an Arm architecture SoC, and that we’re therefore looking at a GD32F330-family offering based on the Cortex-M4 and containing up to up 128 Kbytes of flash memory and 16 KBytes of SRAM, along with a host of peripherals. The family is spec’d to run up to 84 MHz but, considering the 25 MHz crystal further to the right, is seemingly underclocked to its peak potential. Thermal management, don’cha know.
That’s all I have for you today, folks. As always, sound off with your thoughts in the comments!
—Brian Dipert is the associate editor, as well as a contributing editor, at EDN.
Related Content
- SFP: Add-in module delivers diminutive performance and flexibility
- SFP: The single-mode module autopsy
- SFP: The multimode module autopsy
- The multi-gig Ethernet migration: Motivations and implementations
- Design considerations in high-speed fiber networks
The post Disassembling a SFP Ethernet adapter appeared first on EDN.