DDR4 C-die vs D-die: what changes, and why cross-reference matters

The part number stayed the same. The silicon did not.

Memory vendors move a product from one fabrication node to another, correct minor process items, and keep selling it under the same marketing part number. The datasheet headline does not change. The die behind it has. For a design that was validated once and then shipped for seven years, that gap is where intermittent field failures come from.

Die revision is not a secret — it is encoded in the ordering number, if you know which character carries it. Our own cross-reference table makes the point unusually clearly, because we treat the revision as part of the specification and therefore map it to a different HGCHIPS part.

A worked example straight out of our table

Consider the SK Hynix 8 Gbit DDR4 x8 family. Same density, same organisation, same speed bin, same package — and two different die revisions visible in one character of the marking:

The same device family also appears at other speed bins in our table: the C-revision part at DDR4-2933 maps to HGD48G08H29 and at DDR4-2666 to HGD48G08H26. So in the row above, the only difference between the two originals is the revision character — and our mapping gives them separate part numbers rather than treating them as one line.

That separation is a deliberate supply-chain choice, not a marketing one: it lets a customer's approved-vendor list name the exact behaviour that was tested, instead of naming a specification and hoping the next lot matches it.

What a die revision typically changes — and what it does not

The functional contract survives: command set, ball map, density, organisation, package and the advertised speed bin stay put, which is why the vendor can keep the marketing number. What moves is usually at the margins:

• Timing and signal-integrity behaviour. Characteristics you model — input/output behaviour, IBIS-style models, timing margins at the fast corner of your bus — can shift between revisions even at the same rated speed.
• Memory-controller training outcomes. A board with generous training margin will not notice; a board that passed at, say, 6 % margin can pass on one revision and fail on the next.
• Temperature and voltage sensitivity. A shrink changes how the device behaves at the edge of its rated range, which is where industrial designs live.
• Refresh and retention characteristics, which matter for designs that stretch the specified window.
• Package and substrate details, sometimes including the marking style itself — useful when your incoming inspection is trying to determine what actually arrived.

None of this makes a revised part "worse". It makes it different, and different parts need the same evidence to be released into the same design.

Where our DDR4 line stands today

Our DDR4 cross-reference holds 30 parts: densities from 4 to 32 Gbit (8 Gbit is the bulk), speed bins DDR4-2666 (14), DDR4-3200 (13) and DDR4-2933 (3), in x8 and x16 organisations. Where the same original family appears at more than one revision, the table lists them as separate rows with separate HG part numbers — as in the example above.

Adjacent point for low-power designs: our LPDDR4 / LPDDR4X table (39 parts) records the supply-rail scheme explicitly for each part — 1.8/1.1/1.1 versus 1.8/1.1/0.6 versus 1.8/1.2/1.2 — because the rail set is a real electrical difference that a "both are LPDDR4" summary would hide.

What a buyer should actually do about revisions

• Name the revision in the AVL, by full ordering part number, not by the family name. If your AVL says "8 Gbit DDR4-3200 x8", a revision change is already allowed by your own document.
• Ask for the marking and the lot before the reel arrives, and require per-lot records with each shipment. This is the control that catches revision drift as it happens rather than after field returns.
• Define a re-qualification trigger now: new lot, new revision, new package or substrate, new grade. Then run the fast version of it — boot, margin test at your worst temperature, a soak — rather than the full original characterisation.
• Buy a matched second source rather than a mixed market lot. A supplier who can tell you which revision is on the reel is worth more than one who is marginally cheaper and cannot.

If you want to know which side of that line a specific part number sits on, tell us the marking and we will confirm the die-revision position, the equivalent HGCHIPS part, and the test evidence that backs the comparison: send an enquiry or search in the cross-reference tool.