
SD Express Socket Retrofit: Upgrading a UHS-I Footprint Without a Board Revision

The upgrade that is mechanical, not electrical
SD Express is often described as a card-level change: faster cards, PCIe interface, NVMe-like throughput. For the hardware designer the practical question is narrower — can an existing board be upgraded, or does it need a revision?
The answer turns on one thing: whether the socket on the board has the second contact row, and whether the host routed it. The mechanical footprint can be preserved. The electrical routing usually cannot.
This article sets out what survives a retrofit, what has to change, and how to pre-place a board today so the upgrade is possible later.
What the second row is for
An SD card has two possible interfaces behind the same plastic body. The first row carries the classic SD interface up to UHS-I. The second row was added for UHS-II and is reused by SD Express for PCIe.
| Interface | Rows used | Approximate ceiling | Host requirement |
|---|---|---|---|
| Default / High Speed | 1 | ~12.5 / ~25 MB/s | SD controller |
| UHS-I | 1 | ~104 MB/s | SD controller with 1.8 V signalling |
| UHS-II | 2 | ~312 MB/s | SD controller with UHS-II support |
| SD Express (SD 7.0) | 2 | ~985 MB/s | PCIe Gen3 ×1 host plus SD controller |
| SD Express (SD 7.1) | 2 | ~1.97 GB/s | PCIe Gen3 ×2 host |
| SD Express (SD 8.0) | 2 | ~3.94 GB/s | PCIe Gen4 ×2 host |
A single-row socket physically cannot carry PCIe. There is no pin to route the lane to. This is the fact that decides whether a retrofit is possible at all.
The three retrofit cases
| Current board | Socket | Host routing | Retrofit possible? |
|---|---|---|---|
| Case A | Single-row UHS-I | SD bus only | No. Requires a new socket and new routing — a board revision. |
| Case B | Second-row capable, routed to SD controller only | Second row present but not connected to PCIe | Partly. The socket is right; the host is not. Re-spin required for the PCIe traces. |
| Case C | Second-row capable | Second row routed to a PCIe-capable host | Yes. Controller or firmware change only. |
Case C is the one worth engineering for, and it costs almost nothing at layout time. The socket is already the right part; the second row is already routed; the host already has PCIe. The upgrade becomes a question of which controller is populated and what firmware supports.
Most boards are Case A or Case B, which is why the practical advice is not “retrofit” but “pre-place”.
The pre-place strategy
If a product might need SD Express in a future variant, there are five design decisions that cost little now and are expensive later.
| Decision | Cost now | Cost later if omitted |
|---|---|---|
| Specify a socket with the second contact row | Small unit cost delta | New socket footprint and possibly a new board outline |
| Route the second row to the host even if unused | Board area and layer usage | Re-spin |
| Provide ground adjacent to each differential pair | Plane design discipline | Signal integrity problems at Gen3/Gen4 |
| Keep the differential pairs short and matched | Layout effort | Cannot be fixed without a re-spin |
| Thermal vias under the socket shell | Via count | Throttling that cannot be resolved |
The routing decision is the one that gets cut first when a board is space-constrained. Two unused differential pairs running from a socket to a controller area look like waste on a crowded layout. They are cheap insurance against a board revision.
If the traces must be cut for space, an alternative is to route them to a set of test points or a zero-ohm link position, so a future revision can populate the path without changing the layout.
What changes in the host
| Element | UHS-I board | SD Express board |
|---|---|---|
| Controller | SD host controller | SD controller with PCIe, or separate PCIe root port |
| Reference clock | Typically not required for the SD bus | PCIe reference clock required |
| Power | 3.3 V and 1.8 V signalling | Same rails, plus thermal headroom |
| Layer count | Sufficient for SD bus | Often needs more layers for impedance control |
| Firmware | SD stack | PCIe enumeration plus NVMe-style access |
The reference clock is an easy item to miss. A PCIe link needs a reference clock that the SD bus does not, and if the board does not have one routed to the socket area, the retrofit fails on a detail unrelated to the socket.
Layer count is the other. PCIe Gen3 differential pairs at 8 GT/s need controlled impedance and a solid reference plane. A two-layer or four-layer board that was adequate for SD may not be adequate for the differential pairs, and that is a stack-up change, not a routing change.
Signal integrity requirements that scale with generation
| Generation | Lane rate | Practical routing constraint | Board impact |
|---|---|---|---|
| Gen3 ×1 | 8 GT/s | Controlled impedance, minimal stub, solid reference plane | Moderate |
| Gen3 ×2 | 8 GT/s per lane | Two matched pairs, tighter length matching | Higher |
| Gen4 ×2 | 16 GT/s | Low loss material may be required | Significant — stack-up and material change |
Gen4 is where the retrofit argument usually ends. At 16 GT/s the dielectric loss of a standard FR-4 stack becomes the limiting factor over any appreciable trace length, and the answer is a different board material. That is not a socket decision at all.
Gen3 is the realistic target for a retrofit or a pre-place, and it already delivers roughly 985 MB/s over a single lane — a tenfold improvement over UHS-I.
Thermal consequences of the upgrade
A tenfold increase in throughput is a tenfold increase in the heat that has to leave the card. The socket that was adequate at 104 MB/s becomes part of the thermal problem at 985 MB/s.
| Measure | Why it matters after the upgrade |
|---|---|
| Metal shell with large solder tabs | Conducts heat into the board |
| Ground plane area under the socket | The board is the heatsink |
| Thermal vias to inner planes | Spreads heat away from the socket area |
| Airflow or conduction path to the enclosure | Board alone may not be enough |
If the product was designed for UHS-I and never measured card temperature, the upgrade will expose throttling that looks like a performance regression. Measure card surface temperature before and after, in the actual enclosure.
Practical migration path
| Stage | Action | Outcome |
|---|---|---|
| Now, current board | Keep UHS-I; no change | ~104 MB/s ceiling |
| Next revision | Second-row socket, route the pairs to a PCIe-capable host | Pre-placed for SD Express |
| Next revision, minimal | Second-row socket, pairs to test points or links | Upgrade possible without a layout change |
| Upgrade revision | Populate the SD Express controller; update firmware | ~985 MB/s at Gen3 ×1 |
| Later | Gen4 with a low-loss stack | ~3.94 GB/s, at the cost of material and layer count |
The third row is the compromise worth considering for a board that cannot afford the routing now. It preserves the option without consuming two pairs of differential routing across a crowded layout.
FAQ
Can I upgrade an existing UHS-I board to SD Express with a firmware update? No. A single-row socket has no second contact row, so there is no physical PCIe path.
Is the mechanical footprint the same? For microSD Express, yes — it is a drop-in on the same footprint. The change is the contacts, not the outline.
What is the minimum host change? A PCIe-capable controller with a reference clock, plus routing for the second row. Everything else can stay.
Do I need a new PCB material? Not for Gen3. Gen4 at 16 GT/s usually does require a lower-loss stack-up.
Will the socket overheat after the upgrade? It may reach a higher steady temperature because the card is writing faster. Measure card surface temperature in the enclosure rather than assuming.
What is the cheapest way to keep the option open? Specify the second-row socket and route the pairs to test points or zero-ohm links. That costs board area but not a revision.






