Record Details
The Architecture of a Reusable Built-In Self-Test for Link Training, IO and Memory Defect Detection and Auto Repair on 14nm Intel SOC
ScholarsArchive at Oregon State University
| Field | Value |
|---|---|
| Title | The Architecture of a Reusable Built-In Self-Test for Link Training, IO and Memory Defect Detection and Auto Repair on 14nm Intel SOC |
| Names |
Querbach, Bruce
(creator) Chiang, Patrick (advisor) |
| Date Issued | 2015-06-12 (iso8601) |
| Note | Graduation date: 2015 |
| Abstract | The complexity of designing and testing today's system on chip (SOC) is increasing due to greater integrated circuit (IC) density and higher IO and memory frequencies. SOCs for the mobile phone and tablet market have the unique challenge of short product development windows, at times less than six months, and low cost board and platform that limits physical access to test access ports (TAP). This dissertation presents the architecture of a reusable built-in self-test (BIST) engine called converged pattern generator and checker (CPGC) that was developed to address the above challenges. It is used in the critical path of millions of x86 SOC for DDR3, DDR4, LP-DDR3, LP-DDR4 IO initialization and link training. The CPGC is also an essential BIST engine for IO and memory defect detection, and in some cases, the automatic repair of detected memory defects. The software and hardware infrastructure that leverages CPU L2/L3 cache to enable cache based testing (CBT) and the parallel execution of the CPGC Intel BIST engine is shown to improve test time 60x to 170x over conventional TAP based testing. In addition, silicon results are presented showing that CPGC enables easy debug of inter symbol interference (ISI) and crosstalk issues in silicon and boards, enables fast IO link training, improves validation time by 3x, and in some instances, reduces SOC and platform power by 5% to 11% through closed loop IO circuit power optimization. This CPGC BIST engine has been developed into a reusable IP solution, which has been successfully designed into at least 11 Intel CPUs and SOCs (32nm-14nm), with seven of these successfully debugged, tested, and launched into the market place. Ultimately has led to over 100 million CPUs being shipped within one quarter using this architecture. |
| Genre | Thesis/Dissertation |
| Access Condition | http://creativecommons.org/licenses/by-sa/3.0/us/ |
| Topic | BIST |
| Identifier | http://hdl.handle.net/1957/56184 |
