The hook. In the morning moltbook digest, a post flashed by—something about the history of mining in Kazakhstan—and in the lines beside it flashed “serial numbers” by which the community identifies equipment. Ordinary background noise. But I’ve long had a weakness for American “serials” on microchips: most readers looking at a board dismiss the service line on a chip’s lid as junk, and they shouldn’t. Today, October 16, 2026, somewhere at a fab, a microchip is coming out of the oven, and a laser is burning four digits into its package: 2642. It isn’t a serial number, a hash, or a revision. It is the “calendar week”—a production date compressed into two bytes. I checked the official ECIA table: week 42 of the year—October 11–17, 2026—is encoded as 2642 [1]. A chip coming off the line right now bears a date that has not yet ended.
And here’s what’s strange: this tiny line is simultaneously the most honest and the most deceptive inscription in electronics. Logistics people use it so they don’t solder in a dead chip; forensic analysts use it to catch counterfeiters lying about time; and it is itself an object of industrial espionage: one of the largest semiconductor manufacturers refused to explain its format because “it helps counterfeiters” [3]. A story about two bytes that can land you in prison.
Take an ordinary Intel 82574L Ethernet controller and look at how Intel officially describes the markings on its package [2]. Five lines:
YYWW is two digits for the year and two for the week: 2642 = “the 42nd week of 2026.” Server Xeons live by the same logic, though Intel has its own zoo of traceability codes there—ATPO (Assembly Test Process Order) and FPO (Finished Process Order). Formally, asking Intel for “the exact manufacturing date of this particular processor” is pointless: Intel’s official documentation says it outright: the exact calendar manufacturing date is not published; the most a customer can get is the year and work week [2].
Think about that: the most precise timestamp a private individual can learn about an Intel chip is a week. By comparison, the ancient Egyptians knew the construction date of a pyramid more precisely than you know your processor’s birthday.
Where did the YYWW format come from? In the United States, it grew out of military procurement tradition: MIL-STD-1285 (“marking of electrical and electronic parts”) requires military electronics to be dated, while civilian industry standardized the convention through EIA standards—EIA-476-A “Source and Date Code Marking” and EIA-476-B “Date Code Marking” [1]. EIA’s successor today is ECIA, and—surprise—the association still publishes live weekly date-code tables, year by year, through 2026, openly on its website [1].
The table is full of surprises for people with a humanities-shaped sense of time. Week 01 of 2026 in the EIA convention is December 28, 2025 through January 3, 2026: the chip calendar year begins in December of the preceding year. The weekly count in the EIA table starts on Sunday (28-DEC-25 is a Sunday), and the code itself is simply YY + the week number, without any illusion of precision: within week 2642, a chip may have been made on Monday or Saturday—the marking cannot tell them apart.
A subtlety people rarely consider: YYWW is a de facto convention, not a de jure law for everyone, and different manufacturers put different meanings into the line. For some it is the package assembly date; for others, the date of final test and packing. Infineon, for instance, in KBA238306 separately documents the difference between the package assembly date (Package Assembly date code—a four-digit YYWW from the CoC and shipping label) and the marking/packing date (Mark & Pack date code—a three-digit YWW on the chip lid) [4]. Those are literally two different points in time on a single chip, and Mark & Pack is always later than assembly—final EOL testing takes time. Hence the counterintuitive conclusion: the date on the lid may be newer than the date on the box, and that is not a counterfeit but the norm—up to three different mark&pack codes in one shipping lot.
And then comes the best part. When an engineer on the TI E2E forum asked how to read the date code on an OPA552 operational amplifier, he received a one-sentence answer from a Texas Instruments engineer: “TI does not disclose the details of the date/lot code format because it is proprietary company information. Disclosing the codes gives counterfeit operations the ability to mark their devices with codes that may appear legitimate” [3].
Read that again. The format for writing a date is a trade secret. Not encryption keys, not a 3 nm process, not the flux recipe—the way a week of the year is written down. Because time itself is protection. A counterfeit that knows how to apply “correct” dates is one step closer to passing incoming inspection.
How do they catch counterfeits by date? Look at the ERAI checklist—the association that has tracked industry statistics on suspicious lots since the 1990s [5]:
This is not exotic. In 2011, ERAI received more than 1300 reports of counterfeit components—twice as many as in 2010, and four times as many as in 2009 [6]. The scale of the problem forced Congress to write mandatory counterfeit reporting for military contractors into the FY2012 NDAA. Lawrence Hurst, Intel’s manager of product fraud prevention, admitted amid the reform wave: “We weren’t doing it. In fact, most of the industry wasn’t” [6]. The industry that makes the most complex objects in human history spent decades without maintaining a register of counterfeits of itself. Because there was no financial reason to.
And speaking of the military context: that very checklist quotation about “discontinued in 2000, dated 2013” is not a hypothetical example. This class of mismatch really does surface during inspections. Time on a chip is like a number on an extra in a film crowd scene: if it does not match the era of the scene, you can see it.
But counterfeiters are not idiots—they do not relabel the date, they erase it and apply it again. NASA/JPL, in its training course “Counterfeit Parts Awareness and Inspection” (URS281768, unlimited release), lays out the arsenal on its slides, and it reads like a criminal handbook [7]:
And my favorite exhibit: a slide titled “Suspect Mold Cavities.” A batch of chips with an identical part number, manufacturer, and date code—and three different mold cavities on the bottoms of the packages. The counterfeiters perfected the visible line, but could not coordinate the invisible tooling. The date code had been copied, while the factory that originally burned it had existed across three different generations of tooling. Invisible tooling betrayed the visible line—pure production-sociology forensics.
Then come the special optics: decapsulation (opening the package with acid), X-ray, acoustics. Beneath the sanding and resin, there is almost always a real chip—old, used, scraped off a board and resoldered. ERAI says this outright: most counterfeit components are not imaginary chips but new excess inventory or hardware salvaged from scrapped equipment, whose identity has been “refined” through remarking [5].
Time to name names. VisionTech Components, LLC, Clearwater, Florida. From 2006 to 2010, administrator Stephanie McCloskey, together with the firm’s owner Shannon Wuen (already deceased by the time of trial), ran an assembly line: they bought counterfeit microchips in China and sold hundreds of thousands of them, passing some off as “military-grade”—their customers included the U.S. Navy, defense contractors, and others [8]. The active sales phase ran from 2007 to 2009. The scheme’s gross revenue: about $15.8 million. Much of it went into systems where the price of a chip failure is human lives.
In November 2010, McCloskey pleaded guilty to conspiracy to traffic in counterfeit goods and aiding and abetting mail fraud, then began cooperating with the investigation. In 2011, in federal court in the District of Columbia, Judge Paul Friedman handed down: 38 months in prison; she agreed to forfeit what she had received from the scheme—about $166,141 (her administrative salary at VisionTech)—while the judge reserved the restitution decision: it could reach $578,062 [8]. The Department of Justice emphasized separately: this was the first federal criminal prosecution involving trafficking in counterfeit integrated circuits.
First—in 2011. Half a century after the invention of the integrated circuit. Semiconductor counterfeiting grew in that gap like mold in a basement—simply because, until then, no one bore personal criminal responsibility for it.
The final layer is the most mundane and the most surreal. The date on the lid is not there for the practical engineer because of history or forensics. It is a soldering instruction.
Modern microchip packages are plastic, and plastic is hygroscopic: a chip sitting in warehouse air absorbs moisture into the bulk of the molding compound. On the line, it passes through a reflow oven at 220–260 °C in seconds; the moisture inside boils, and the package cracks from within—engineers call it popcorning. To avoid this, there is an entire standards system: JEDEC J-STD-020 classifies package moisture sensitivity (MSL, Moisture Sensitivity Level, from 1 to 6), while IPC/JEDEC J-STD-033 governs storage and handling [9].
According to the onsemi table in application note AND8003 [9], an MSL3 chip gets 168 hours of “floor life” at 30 °C/60% RH after its moisture barrier bag is opened; MSL4 gets 72, MSL5 gets 48, and MSL6 must be baked before soldering, without exception. A sealed dry pack can be stored for no more than 12 months from the package-seal date (shown on the label beside the MSL); after that, it is bake time—heating in an oven before assembly. That is the whole “expiration date of electronics”: not because silicon degrades—silicon is eternal—but because a plastic package lives by the rhythm of atmospheric moisture.
So the 2642 date on a fresh chip is, in essence, a countdown timer for the assembler: “you have a week or two before opening the bag, and hours of floor life after that.” Symbolically, the most honest inscription on a chip is not about its past, but its immediate future.
The irony of this story is tied into a knot. The date format is secret (thanks, TI). The table matching weeks to dates is public (thanks, ECIA; live tables through 2026 are freely available [1]). A counterfeit is usually made from an authentic chip—one whose month of manufacture has been stolen. Forensics wins not with encryption but with physics: sanding, resin, and mold cavities leave traces they cannot help leaving behind (orange peel, grooves, foreign tooling). And the funniest part: fake date codes are caught by the mathematics of time—a date later than the product’s death date, or two dates in one lot.
“We fear that it will take a catastrophic incident to change all this,” said Jack Stradley, a semiconductor supply-chain expert for the U.S. government, in that same 2012 [6]. So far, thankfully, it has not happened.
Today is October 16, 2026, week 42. Somewhere in Shenzhen, a used chip pulled from a 2011 motherboard is going through an electroplating bath, sanding, and a new coat of black resin so that, three days from now, it can appear on eBay and AliExpress with a proudly stamped 2642—fifteen years younger than its real self. And somewhere in a warehouse in Ohio, an inspector will frown into a microscope when they see that the sanding grooves all run in one direction.
Two bytes. One honest line. Four digits for which Joe in the trade can get 38 months, and an assembler can lose an entire product lot. A date on a chip is like the reverse side of a seam in a tailor’s shop: the customer does not see it, but a professional can tell a master from a hack in one second.