Quality by Design

Enhanced Design-for-Reliability (DFR) methodology and guidelines have proven a FIT rate less than 12 at production. Appropriate reliability design rules, budgets and guardbands were established to address post-stress degradations and long-term reliability.

AMD reliability methodologies overcame the shrinking reliability margins for the leading process nodes. By leveraging early learning, in-depth tools expertise and machine-learning, AMD engineers shortened development processes from months to days, and accommodated the extra iterations required for 28/20/16nm. As a result, AMD devices are meeting the stringent requirements of the most reliability-sensitive applications in industrial, automotive, aerospace, and defense industries through:

• Improved outlier elimination and new DFR methodology are combating the “shrinking bathtub” curve
• AMD engineering and quality assurance programs have yielded proven, predictable, and very-low failure rates over extended device lifetimes
• The 28/20/16nm devices were released to production with power and defect density (DD) beating previous estimations

Figure 1. From 28nm to 16nm, wafer-level reliability exceeds transistor and interconnect market requirements to deliver industry-leading device FIT.

“If you can’t test it, then don’t design it.” AMD design engineers place priority on the ability to detect problems quickly and identify root causes that enable solutions. The AMD approach starts with Design-for-Test (DFT) methodologies and the New Product Introduction (NPI) process, and extends throughout the complete product life cycle.

AMD continuously improves the test methodology of all products from generation to generation through implementing various Design-for-Testability (DFT) methods and tools. These techniques span digital logic, IP, memory elements, I/O boundary scanning, and many other areas. New test methodologies are measured against PPM results from customer returns. It builds quality into the design of world’s most advanced technology and largest All Programmable / adaptive SoC die.

Design methodology focused on quality integration through all phases of design cycle:

• DFX key features such as PS core LBIST / MBIST / JTAG / ECC / Redundancy
• Validation platforms emulation and sub-system
• Extensive use of test chips to validate key IPs (DDR / SERDES / MC)
• Extensive use of Open Source community for System Software Solutions (PetaLinux)

Table 1. Key Manufacturing Tests: These test examples illustrate the test methodology innovation and continuous improvements that are intrinsic parts of the AMD engineering culture. Learnings from previous generations and excellent coverage are leveraged to overcome the challenges with newer technology and ultimately enable the release of high-quality adaptive SoCs.

AMD Design-for-Manufacturability (DFM) engineering discipline ensures quality, reliability, and time-to-market by focusing on mitigating risks and optimizing operational excellence. It starts with rigorous design of experiments, tests and manufacturing methodologies such as:

• DFM rules that provide performance advantages
• Modular design and reuse with technology risk proofing & mitigation
• Characterizing devices more deeply using advanced tools & methods
• Reliability prediction, using a custom-hierarchical, end-to-end tool that draws on an AMD design database
• Expanded wafer qualification using three additional elements for assembly testing: electrical, thermal, and mechanical
• Ensuring multiple-die performance by adding system-timing checks to characterization process
• Enabling power die optimization during selection, through fully-integrated die-testing at wafer sort, now standard across the 7-series, UltraScale, and UltraScale+™
• Collecting more data to prove operational performance & software models

These methodologies behind the industry’s first 28nm (28Gb/s) 3D IC heterogeneous devices and now extended at 20nm (33Gb/s) and 16nm (56Gb/s) – Figures 1 and 2 – are fully integrated to meet performance and specifications.

AMD New Product Introduction (NPI) process was established and refined over the last five generations of AMD technology, culminating in our 28/20/16nm products, which include our FPGAs, 3D ICs and SoCs devices. The NPI is a robust and rigorous process with stringent exit criteria targeting the highest quality of AMD products in Engineering Silicon (ES) and production release phases. Some NPI activities include:

• Performance validation over process corners, to check for margin over PVT for manufacturability
• Comprehensive attributes, with sweeps above and below default settings to confirm margin and highlight sensitivities to PVT, with higher focus on applicable MGT, BRAM, and MMCM
• Improved dynamic power characterization to better isolate specific power blocks, with extensive validation of Xilinx Power Estimator (XPE) and Xilinx Power Analyzer (XPA) models over toggle rates and PVT
• NPI process integration with product life cycle

Figure 1: From first silicon to production material ship, AMD has redefined its verification and characterization process to drive early discovery to release with zero errata.

The results are in and have ended all debate. The company embraced a hallmark zero-defect mentality with focus on quality and reliability-which spans across design, process development, assembly, and test-guided by fifth-generation new product introduction (NPI) processes and the most stringent release criteria to date.

The end-result of AMD commitment to quality is exhibited with Customer Quality Scorecards, a direct feedback mechanism provided by our customers. With all the challenges related to shrinking process nodes, DFT, DFR and DFM expectations keep increasing to sustain the excellent results, demonstrated by customer-experienced PPM results.

AMD Verification & Characterization (V&C) ensures that products operate over the specified voltage and temperature ranges. V&C starts with building in sufficient margin in the design phase, which is then validated during simulation prior to product tape out. Critical circuits are verified using test vehicles. Finally, production-ready product quality is characterized based on the datasheet and AMD IP across all process corners, voltage, and temperature (PVT) combinations. The data is used to evaluate product performance against the published datasheet and established production test margins.

AMD engineers have proven that they verify, characterize, test, and qualify devices better and faster than anyone else in the industry. At 16nm, the advancements improved data collection, verification, and characterization with:

• Highly automated design flows and timing analysis
• Verification and Characterization (V&C) process for earlier identification of issues using earlier corner material
• Known-good die testing at wafer sort, which is critical due to 3D slice density and yield
• Die selection, power die optimization, and performance to address the most stringent requirements (power vs. performance)
• Test methodologies for processor subsystem (PSS) and FPGA interaction, to meet performance, specification, and quality requirements
• Expanded data collection and test cases for verification and characterization (including system-level test)
• Joint-safe-launch quality at the system-level for advanced systems and packaging to track more advanced applications, with system-level testing being a significant component in the AMD Volume-System-Testing (VST) over 10,000 devices to emulate customer use cases

Figure 1. SSO Data: With package complexity, SSO characterization data validates AMD simulation methodologies that contribute to packaging design.

With every product release, AMD publishes characterization reports that validate product datasheet specifications and give customers a better understanding of AMD methodologies and techniques. The following steps depict the key milestone criteria prior to release of characterization reports and product specifications:

• Test patterns for all blocks, during both automated and bench testing
• Full-functionality verification across speed, process, and temperature
• Power and performance verifications, corner cases, and system testing
• Volume automated-test-equipment (ATE) characterization

Figure 2. Sample Characterization Report: With every product release, AMD publishes characterization reports that validate product datasheet specifications and give customers a better understanding of AMD methodologies and techniques.