Timeline for Designing a U.2 to M.2 NVMe adapter
Current License: CC BY-SA 4.0
9 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Apr 18 at 4:46 | comment | added | alex.forencich | Don't forget a 1:4 clock buffer for the ref clock | |
| Mar 16 at 20:22 | comment | added | Polynomial | @Ale..chenski Unnecessary for this one-off. The cables are a known quantity - they're specifically designed for PCIe with nominal 85Ω. The connectors are designed and specified for PCIe 4.0, which is twice the Nyquist rate I'll actually be running at. The board stackup is controlled impedance and I have the EDA tooling to do field sims and verify timing skew and intrapair phase skew along the length of the trace. It'll be around 25mm of trace on the board max, which translates to under 1dB of loss (accounting for dielectric loss + conductor loss + estimated 6μm roughness loss). | |
| Mar 16 at 18:08 | comment | added | Ale..chenski | All sounds good, but how do you plan to check the characteristic impedance of your channels across cable-connector1 - trace - connector2 pass-through? Do you have 2-channel differential TDR or 4-channel LNA equipment to validate quality of your layout? Do you know the cost of this equipment? | |
| Mar 16 at 6:25 | history | edited | Polynomial | CC BY-SA 4.0 |
minor edit re: timing skew
|
| Mar 16 at 6:24 | comment | added | Polynomial | @Ale..chenski But, as I said in the post, it's not just about the money. I don't trust the design quality of those commercial adapters at all. Sure, they'll function, but these things are designed cheap and sold for a high price because they're niche. I wouldn't trust that they used high reliability caps with decent margins on ripple current rating, or that they did a good job of the DC-DC layout, or that they made any effort to minimise conducted noise. The physical footprint is also less than ideal and I'd really prefer to have everything on a single ~100x100mm card. | |
| Mar 16 at 6:16 | comment | added | Polynomial | @Ale..chenski Yes. The price of the four commercial adapter cards and the traditional U.2 cables comes out at about £100 (GBP). The price of a custom 4L impedance controlled PCB is about £9.75 (for 5pcs), the multiphase buck IC is £12, the power inductors are another £2, caps and passives are under £1. The U.2 connectors are roughly £12.50, another £5 for M.2 connectors. I already have the required cables (Mini-SAS HD to Mini-SAS HD) spare from a RAID card, but even if I didn't it's about £15-20 for a pack of four. Add it all up, and it's still two thirds the price of the commercial approach. | |
| Mar 16 at 5:44 | comment | added | Ale..chenski | I am wondering, do you have a ballpark estimate of the cost of your design, including PCB development ("less amateurish"), manufacturing, cost of connectors (m.2 and PSAS4M2130081TR), all other components (at single qty pricing), etc.? | |
| Mar 16 at 3:34 | comment | added | Polynomial | This is a personal design, and sadly PCI-SIG membership pricing is way beyond my means. If you happen to know the specific document names or identifiers, though, I may be able to acquire them via friends. | |
| Mar 16 at 3:07 | history | asked | Polynomial | CC BY-SA 4.0 |