
企业会把注册过的ECC DDR5内存条交给维修店尝试修复,而不是直接丢弃坏件换新,这绝对是时代的标志了。[northwestrepair]就因此拿到了两根ECC RDIMM,按当前市场行情估值够买辆不错的入门级汽车。
128GB那根先搬上工作台,故障表现为开机时DDR5内存训练失败。这说明一个或多个DRAM BGA封装可能存在焊点开裂,理论上经过更多诊断测试后,可以对单个IC重新植球修复。
这次只需做一个回流焊接周期就让内存起死回生,DDR5内存训练在测试平台上一路顺利通过了。
256GB那根就没这么好修了,完全没反应。这说明PMIC(电源管理集成电路)坏了,因为DDR5内存上负责给DRAM芯片供电的就是这个元件。另一个可能是SPD ROM出问题,所以先换了它,这需要重新编程RDIMM。之前用dump下来的SPD镜像刷写,但换芯片毫无作用。
之后又对几个可疑IC做了多次回流焊接,依然毫无反应。接着进行了大量重新植球和其他故障排查操作,结果问题是一个短路的MLCC(多层陶瓷电容)加上损坏的PMIC,很可能是有人把这RDIMM摔到服务器机房地板上导致的。
MLCC开裂确实很常见,因为它们常被用作滤波电容,这款RDIMM就是这样。多检测几个此类MLCC是否存在短路,是缩短诊断流程的好方法。由于这根RDIMM曾被摔落,把所有IC都重新植球应该还是明智之举,因为这可能导致部分焊球开裂。
本文来源:https://hackaday.com/2026/10/07/repairing-a-couple-of-very-expensive-registered-ddr5-ram-sticks/
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
[[IMG0]]Companies sending registered ECC DDR5 memory sticks to repair shops for attempted fixes rather than simply discarding them and buying replacements is surely a sign of the times. [northwestrepair] obtained two ECC RDIMMs this way, valued at current market prices enough to buy a decent entry-level car.
The 128GB stick went first. The fault manifested as DDR5 memory training failure on boot, indicating one or more DRAM BGA packages may have cracked solder joints. After additional diagnostic testing, individual ICs can theoretically be reball repaired.
In this case, a single reflow cycle brought the memory back to life, and DDR5 memory training passed smoothly on the test platform.
The 256GB stick was less cooperative—it was completely unresponsive. This pointed to a failed PMIC (power management integrated circuit), since that's the component responsible for powering the DRAM chips on DDR5 memory. Another possibility was an issue with the SPD ROM, so that was swapped first, which requires reprogramming the RDIMM. A dumped SPD image was used to flash it, but replacing the chip made no difference.
Multiple reflows on several suspicious ICs followed, still with no response. After extensive reballing and further troubleshooting, the issue was found to be a shorted MLCC (multilayer ceramic capacitor) combined with a damaged PMIC—likely caused by someone dropping the RDIMM on the server room floor.
MLCC cracking is indeed common since they're often used as filter capacitors, as is the case with this RDIMM. Testing a few more such MLCCs for shorts is a good way to streamline the diagnostic process. Since this RDIMM was dropped, reballing all the ICs was still the prudent move, as the impact could have cracked some solder joints.