Whose DRAM chips are actually inside your memory stick

There is an asymmetry in this market worth naming at the outset, because it shapes every section below. The chip makers have no commercial reason to tell you which modules contain their dies, and the module brands have a commercial reason not to.

On the chip-maker side, SK hynix’s own newsroom says it "is classed as an integrated device manufacturer (IDM) that is capable of handling both the design and production of chips at its in-house facilities" 5 — a description of what it makes, not of who resells it. Checked for a published list of module brands using its dies: its brand-identity page 6, its industry-players explainer 5 and its back-end process series 7. No such list was found on any of them.

A concrete instance of that asymmetry. Lexar has published an "Empowered By SK hynix" Gold Label certification, dated 2023, on at least one product page — recorded and sourced on our Lexar guide, and not present on the THOR OC page checked here 8. A targeted search of skhynix.com, news.skhynix.com and product.skhynix.com for any SK hynix side of that programme returned nothing, and SK hynix’s brand-identity page carries no logo-use, co-branding or partner-certification guidance 6. So the claim that a given module holds SK hynix dies is made by the module vendor and cannot be checked against anything SK hynix publishes. That does not make the claim false. It makes it unverifiable from the outside, which is a different and more common situation than it sounds.

One important limit on this whole page, stated here rather than buried. SK hynix’s entire product documentation was unreachable for this check. Its corporate site publishes no product pages and routes all product content to product.skhynix.com, whose robots.txt returned an HTTP 500 on one attempt and timed out on another, so our retrieval tool declined every path on that host — including its part-number search tool and all its datasheets. That is our fetcher’s policy on an unreadable robots file, not a refusal by SK hynix. Where this page reports that SK hynix publishes nothing on a question, read it as “not establishable from here” rather than as a finding about SK hynix. Samsung and Micron were both readable, and are treated accordingly.

Samsung publishes a part-number decoder too, and the story of it is worth telling because it is a case where a company’s navigation and its file server disagree.

Samsung’s support section does not link a part-number decoder. Its support index lists Quality Support, Tools & Resources, Consumer Storage Support, FAQs and contact information, and no decoder appears among them — nor on its Tools & Resources page, which lists Technical Resources, Fabrication Process, Dictionary, DS Open-Ecosystem, Magician and Tools & Software 13. But a decoder is live on Samsung’s download host: a PDF titled “DDR3 Code Information”, subtitled “Module” and dated “Aug.2012”, which walks the module part number position by position 14.

Its most relevant field for this page is field 10, “Component Revision”, which enumerates eight die letters, M-die through G-die 14. Samsung also encodes the speed there, at fields 15 to 16, as codes such as "K0 : DDR3-1600 (800MHz@CL=11, tRCD=11, tRP=11)" 14. So on a Samsung-branded module the die revision is not merely knowable — it is a character in the part number.

Three limits on that, all of which matter. First, it is DDR3 and dated August 2012. Probes for DDR4 and DDR5 equivalents on the same host returned 404, and a targeted search across Samsung’s three semiconductor hosts for other “Code Information” documents returned only this one file. Second, it decodes Samsung-branded modules, not a third party’s module built from Samsung chips — it tells you nothing about a Corsair or Patriot stick. Third, it is unlinked: reachable by direct URL but from no navigation path found in this check, and the older www.samsung.com decoder URLs now redirect to the semiconductor homepage. Treat it as a surviving document rather than a maintained one.

There is a sharper Samsung asymmetry on the same question, between how Samsung documents a module in a PDF and how it documents modules on the web. Its DDR4 Unbuffered DIMM datasheet publishes an ordering-information table that names the exact component inside each module — part number M378A1K43DB2-CTD is listed with a component composition of "1Gx8(K4A8G085WD-BCTD)*8" 15. Its public DDR chips-and-modules pages give capacity, speed, voltage and package and no component part number or die revision 1. The information exists, published by Samsung, in the PDF and not on the web page.

A limit on that comparison, stated rather than glossed. Samsung’s individual module product pages could not be retrieved for this check — the registered-DIMM URL tried here resolved to the DDR category page instead, and the part number did not appear on what rendered 16. So what any one Samsung module page does or does not list is not established here, and the contrast above rests on the category pages that did render. The two documents are also different module families: the datasheet covers unbuffered DIMMs, the product URL a registered one.

SK hynix does not publish a marking guide that this check could find, but it does publish the clearest explanation of why anything is printed on a DRAM package at all — and the answer connects markings to testing rather than to branding.

From its own engineering series: "When testing is completed, the test results must be recorded on the exterior of the package and, consequently, laser marking is required." 7 The marking, in other words, is the test result made physical. The same article sets out what is tested and when: "Wafer testing is the process of examining the electrical characteristics of a chip at the wafer level", and "Wafer tests inspect the properties and qualities of the numerous chips that are on a wafer." 7

It also publishes the order of operations: "Wafer testing usually follows the following process: EPM (Electrical Parameter Monitoring) → Wafer Burn-in → Test → Repair → Test." 7

What was checked, and what was not. The testing article and the industry-players explainer were both read; neither contains a guide to what the marking fields mean, and a search of the newsroom found no marking-specific article 7 5. SK hynix’s datasheets, which would carry ordering-information tables, live only on the host that could not be reached at all (see the Expert level of the first section). So the correct statement is that no SK hynix marking guide was reachable, not that none exists.

The practical upshot of this section is narrower than it first looks. Reading a chip marking works well for Micron, because Micron publishes a decoder and prints an MT prefix 17. For Samsung it works if you already know the component part-number scheme, which Samsung publishes in datasheets rather than as a marking guide 15. For SK hynix, nothing readable here supports it. And in all three cases you must be able to see the chips — which on most consumer kits means removing a heat spreader that is often glued on, and doing so may well end your warranty.

Crucial does have one thing settled, though by construction rather than by disclosure. Micron describes Crucial as "Micron’s global consumer brand", "uniquely able to connect millions of customers to the innovation and technology that Micron has been perfecting for more than four decades" 26.

A precision point, because it is tempting to overclaim here. That page establishes that Crucial is Micron’s brand. It does not state that Crucial modules are built with Micron DRAM, and no page was readable in this check that says so in Micron’s or Crucial’s own words — the one page whose title promises exactly that is on the blocked domain. So “a Crucial module holds Micron dies” is a reasonable inference from the brand relationship, and this page presents it as an inference rather than a quotation.

That inference is also now historical at the consumer end. On 3 December 2025 Micron "announced its decision to exit the Crucial consumer business, including the sale of Crucial consumer-branded products at key retailers, e-tailers and distributors worldwide", with shipments continuing "through the consumer channel until the end of fiscal Q2 (February 2026)" 27. Its stated reason, from EVP and Chief Business Officer Sumit Sadana: "The AI-driven growth in the data center has led to a surge in demand for memory and storage." 27 Micron committed to "continued warranty service and support for Crucial products" and to continuing "the sale of Micron-branded enterprise products to commercial channel customers globally" 27.

Which leaves a neat irony worth stating. The one consumer brand whose die maker was never in doubt is the one leaving the consumer market — and it left because the dies were worth more sold to data centres than sold to you. That is the same force behind the substitution clauses above: when die supply is tight and priced by the hour, a brand that fixed its part number to one supplier’s silicon would be unable to ship.

Two notes on how far the admissions go. Corsair’s applies to the ICs specifically and is paired with a code that lets you check after the fact 9. Kingston’s is broader — "different combinations of components", not only DRAM 20 — and is not paired with a consumer-line code that lets you check, since FURY part numbers carry no manufacturer field 12. Of the two disclosures, Corsair’s is the one a buyer can act on.

Now the part that surprised this check, and that corrects a premise we started with. The paywall in this story is in the wrong place.

The documents describing the module are free. JESD400-5D.01, "DDR5 SERIAL PRESENCE DETECT (SPD) CONTENTS", Release 1.4, published January 2026, is listed by JEDEC as "Free download. Registration or login required." 31 So is the SPD hub standard, JESD300-5B.01, published May 2023 32. The only priced document in the chain is the one describing the die: JESD79-5D, DDR5 SDRAM, version 1.41, November 2025, at a non-member price of US$423 33.

So the obstacle to answering “whose dies are these?” from the standards is a registration form, not a paywall. Ten JEDEC pages and press releases were read for this check, and none of the free, unauthenticated material names a manufacturer-ID field in SPD 31 32 34. That is a statement about what the free pages say, not about what the standard defines. Given that Micron labels two entire 64-byte blocks "Manufacturing information" and points at JESD400-5 for their contents, the likelier reality is that the standard does define such fields and the free material simply never describes them. This page asserts neither.

JEDEC does run a registry of manufacturer identities, which is the closest thing to a definitive answer that exists. JEP106BP, "Standard Manufacturer’s Identification Code", published September 2026, defines the format: "The manufacturers identification code is defined by one or more 8 bit fields, each consisting of 7 data bits plus 1 odd parity bit", and it is "assigned, maintained and updated by the JEDEC Office" 35. Note “manufacturers” without an apostrophe in JEDEC’s description, against the apostrophe in its title; both are reproduced as printed.

Crucially, JEP106 is not only a format — it is the register of assignments. JEDEC’s order form states that for a company obtaining a code, "the company name and ID code assigned will be included in JEDEC publication JEP106 (all future versions)", and directs readers to "Refer to JEDEC Publication JEP106" for the list 36. Its own dictionary confirms it: an MFR code is "A code assigned by JEDEC that identifies manufacturers. JEP-106 is the publication containing the codes." 37 Getting one costs "$750 (one time fee)", and "JEDEC Membership is NOT required to purchase an ID Code" 36.

So the code-to-company mapping is published, free, and sits inside a registration-gated download — with no separate free public list found across the three JEDEC pages read here 35 36 37.

One route this page deliberately leaves open rather than closing. DDR5 modules carry an SPD hub and a power-management chip, both standardised 32. Whether either reports a manufacturer identity readable by the host is not established: the register maps that would settle it are inside JESD300-5B.01 and the PMIC standard, and the vendor datasheets that would show them were either behind login walls or truncated before the register tables. Do not read this section as saying they do not. Read it as saying nobody in the free chain prints the byte.

SK hynix comes closest to describing the mechanism, and stops just short of the grading step. Its back-end process series gives the test flow — "Wafer testing usually follows the following process: EPM (Electrical Parameter Monitoring) → Wafer Burn-in → Test → Repair → Test" — and explains that results are then laser-marked onto the package 7.

Note what that article does not contain, because the temptation to fill it in is strong. Checked for the terms EDS, electrical die sorting, speed bin, grading and good-die sorting: none appears. ("Sort" and "sorting" do occur, but only of defects — "manufacturers sort out pressed balls, missing balls, and other defects.") The article describes testing and repair; it does not say how a tested die is assigned to a speed grade 7. The step that connects “we measured this die” to “therefore it ships as DDR4-2666” is published by none of the three makers in anything read for this check.

Nor does JEDEC fill the gap in its free material. JESD79-5D, which defines the DDR5 device, is the priced document at US$423 33. The free material confirms only that bins exist and are enumerated: JEDEC’s 2022 SPD release says "Example SPD codes are given for all JEDEC standard speed grades, assisting suppliers and users in the interpretation of the content definitions", and that the update "covers example timing tables for DDR5 speed bins up to and including DDR5-7200" 34. How a bin is constituted, and how parts are graded into one, is not in the free material.

What the module brands say about their own sorting is thinner than their marketing implies, and one case is worth recording precisely. Corsair’s Dominator Platinum RGB DDR5 product page (CMT64GX5M2B6000C30) carries the feature heading “HAND-SORTED, TIGHTLY-SCREENED MEMORY CHIPS” 39. That page was fetched three times to find the explanatory text beneath it, and there is none — though four of its other feature headings are equally bare, so this is the page’s layout rather than a silence chosen for this claim. Corsair states the screening and states no method. Patriot does the same thing more briefly: "High-quality IC: Selected and screened for enhanced overclocking capabilities", with no chip maker named anywhere on its site 22.

Kingston is the exception, and describes its qualification of suppliers rather than of individual dies: "Because one of the most important components of a memory module is the DRAM, Kingston strictly qualifies our DRAM suppliers", explaining that "Slight physical variances among component manufacturers make this a necessary step", and naming an outside lab — "Kingston uses services provided by AVL for DRAM qualification", where "AVL (Advanced Validation Labs) is a technology services company specializing in the pre-production and post-production testing and validation of memory to industry OEMs" 40.

A contradiction in candour worth noting, between two Kingston pages. The formal corporate testing page says Kingston "strictly qualifies our DRAM suppliers" and names none of them 40. The company blog names all three: "Micron, SK Hynix, and Samsung" 3. The blog is the more forthcoming document.

Kingston’s own pages are the most informative material on this question and they also contain the sharpest contradiction found in this check. Both of the following are on the same support page, about the same practice.

Under the question about adding a module of the same part number later, answer reference KTF-001002-005: "Mixing modules, even if they are the same specification and part number, may result in degraded performance, instability, or incompatibility." 20

Under the question about kit versus individual part numbers, answer reference KTM-020911-GEN-19: "Kingston ensures that only modules with matching components are packaged into dual (K2), quad (K4), and octal (K8) channel kits. Single modules purchased at different times may feature different components. While it is unlikely to cause any performance or compatibility issues, it’s recommended to always buy kits for multi-channel systems, so that all components are the same." 20

So Kingston tells the same buyer, on one page, that mixing same-part-number modules “may result in degraded performance, instability, or incompatibility” and that it “is unlikely to cause any performance or compatibility issues.” Both quotations are Kingston’s, both were retrieved twice, and this page does not attempt to reconcile them. The advice they arrive at is the same; the risk they describe is not. If you are deciding whether to risk it, note that the more cautious sentence is the one attached to the question you would actually be asking.

Kingston also draws a distinction the other brands do not, between standard and overclocking parts: "For JEDEC spec (industry standard) memory modules (ValueRAM, Server Premier, non-overclock), mixing across banks is okay, but Kingston does not recommend mixing memory within the same bank group. Doing so on PCs/laptops disables channel architecture optimization, forcing the memory to perform at half the potential bandwidth." And separately: "Mixing overclockable (Kingston FURY™) memory modules or kits featuring different DRAM densities or specifications in a PC/laptop is not recommended, as this may cause instability or fail to overclock." 42

That distinction is the tidiest summary of the whole page. A JEDEC-standard module is defined by the standard, so mixing is a question about the standard. An overclocking module is defined by a tested profile on particular silicon, so mixing is a question about the silicon — which is the thing you cannot see, whose maker most brands will not name, and which may have changed since the last one you bought.

Software will not rescue you here either. Checked what each brand’s own utility claims to show: Corsair’s iCUE page offers to "Manage DRAM Profiles" and makes no claim to report the IC manufacturer 43; Kingston’s FURY CTRL is described as software to "customize the lighting effects of Kingston FURY RGB products" 44. No vendor tool examined in this check reports the DRAM die maker to the user. Corsair’s route to that answer remains a sticker on the module.

Finally, the shape of the answer, since the individual findings can obscure it.

The information exists at every level except the one you can see. The chip maker knows exactly which die it shipped; it marks the package with the test result 7 and publishes a decoder for the marking, at least in Micron’s case 17. The module brand knows which components it fitted; Kingston says it "strictly qualifies" its suppliers and uses an outside lab to do it 40. The module itself carries two blocks of a chip labelled "Manufacturing information" 28. JEDEC maintains a registry mapping identification codes to company names 35 36.

Every link in that chain is documented. The chain simply is not joined up for the buyer — and nothing read here states why. No party in the chain publishes a rationale for withholding the die maker, so what follows is this page’s reading rather than a sourced finding: a brand tied to one supplier’s silicon loses flexibility when that supplier reallocates output, and reallocation is demonstrably happening. Micron announced on 3 December 2025 that it would exit its Crucial consumer business, with sales continuing to February 2026, "in order to improve supply and support for our larger, strategic customers in faster-growing segments", after "AI-driven growth in the data center has led to a surge in demand for memory and storage" 27.

So the practical answer, ranked by how much work it takes:

1. Buy Corsair and read the version code — V3 Micron, V4 Samsung, V5 Hynix, published by Corsair itself 9. This is the only route on a consumer kit that needs no tools.
2. Buy a Kingston Server Premier, Design-In or ValueRAM DDR4 part and read the manufacturer character in the part number 12. Not available on FURY or on ValueRAM DDR5.
3. Look at the chips. On Micron silicon the five-digit code decodes to a part number through Micron’s own tool, and Micron says it "is typically preceded by an MT mark" 17. The tool is interactive and was not exercised from this build environment; only its description was read. This usually means removing a heat spreader, which may end your warranty, and on Samsung requires knowing the component scheme from a datasheet rather than a marking guide 15.
4. Buy a Samsung-branded module, where the component part number is published in the datasheet’s ordering table — for the unbuffered parts, "1Gx8(K4A8G085WD-BCTD)*8" 15 — while the web pages carry module specifications only 1.
5. Accept that you cannot know, which is the honest position on Lexar, Patriot, PNY, Silicon Power and Mushkin, none of which names a die maker anywhere this check could read 21 22 23 24 25.

And the one piece of advice that follows from all of it, which every brand that addresses the subject agrees on even where it disagrees about the risk: buy the memory you want in a single kit, in one box, at one time 20 10 22 8 23.