Repository navigation
Conversation
|
@CL-BAF is attempting to deploy a commit to the Yoseph's projects Team on Vercel. A member of the Team first needs to authorize it. |
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
|
As discussed on Discord, before the final review you'll:
Once that's done comment here in the thread and I'll take another look and review the PR. |
* Add a DDR5 memory dive (PR #18, rebased onto main) Squashes the eight commits of pull request 18 onto the current main: module, package, bank array and bank scales, with sources and tests. * Rebuild the memory dive and add DDR4 beside it The module, package and diagrams from #18 were rebuilt against data sheets: - Module: one extruded board outline with a real key, latch notches and, on DDR4, the curved contact edge; 288 contacts at the JEDEC 0.85 mm pitch; the published height and thickness with the spreader; PMIC, inductors and SPD hub on DDR5, SPD EEPROM and fly-by termination on DDR4; separate front and back plates with locking clips and pads. - Package: the board-on-chip FBGA both generations use, die face down over a slot in the substrate, centre pads, bond wires, encapsulant and the 78-ball 13 x (3 + 3) grid it leaves room for. - Die: bank groups with their own I/O, 32 banks (DDR5) or 16 (DDR4), and the centre-stripe periphery: command, mode registers, refresh, on-die ECC or DLL, prefetch, data I/O and ZQ. - Bank: mats, local sense-amplifier stripes, sub-wordline drivers, global wordlines and bitlines, decoders and bank control. - New cell scale: wordlines, bitlines, access transistors, charged and empty capacitors under the plate, precharge, sense amplifiers, column select and local I/O. The DDR4 module (FURY Beast DDR4-3200) sits beside the installed DDR5 one in the Memory menu, each in its own dropdown, and the pair forms a Memory group in the comparison workbench. The motherboard now carries the complete DDR5 model in its slots, keyed from the module's own key. * Take DDR4 down to the bank and the cell, and clean up the spreader DDR4 now dives as deep as DDR5. Its bank and cell scales are the DDR5 ones under ddr4 ids with the numbers that differ swapped in: a 17-bit row address and 131,072 rows per bank, 64 bits per column access, and wordlines boosted to 2.5 V instead of 1.8 V. The comparison sheet gains a Bank and cell row saying the same. The spreader and pads no longer flicker as the camera moves. The interlock and the pads shared faces with the plates and with each other; they now stop short of them. The locking clips sit inside the plates' outline instead of standing off the edge as tabs, the wordmark is printed on the plate rather than floating at ridge height, and the back tape fills the gap to the laminate and leaves downward with the back plate instead of rising through the board. --------- Co-authored-by: cameron lonnie <goone499@gmail.com>
|
Thanks @CL-BAF. This work is now on main and live on pc-anatomy.com. #27 squashed your eight commits onto current main with you credited as co-author, then rebuilt the module, package and diagrams against data sheets (Kingston FURY Beast-based DIMM, board-on-chip FBGA, die, bank and cell) and added DDR4 alongside DDR5. Since everything here is already in main, I'm closing this one in favour of #27. Your groundwork made it possible. |
Closes #5
Summary
Automated verification
npm run checknpm run lintnpm test— 61/61 passingnpm run buildgit diff --checkBrowser QA
Ready for maintainer review. The Vercel status remains an external deployment-authorization failure; CI/CD is maintainer-owned and was intentionally left unchanged.