Session 6 AND, AT THE WAFER LEVEL For many in the industry, performing final test at the wafer level is still a novel idea. While providing some much needed solutions, it also comes with its own set of challenges. The four papers in this session look at wafer-level test from a number of different perspectives. The first one discusses the mechanical and electrical differences between wafer-level probe and wafer-level test using spring pins, focusing on requirements for performing final test at the wafer-level. The second presentation provides a comparison between traditional probe test for an RF wafer level chip scale package (WLCSP) and a final test socket solution. TSV issues lead our third author to share technologies that can bridge between 3D stacking and the 3D IC without TSVs. Finally, we ll gain insight into what some consider the holy grail of burn-in and test wafer-level burn-in (WLBI). Now that WLBI is possible, it s important to understand when it s appropriate to consider WLBI versus other burn-in alternatives. Spring Probes and Probe Cards for Wafer-Level Test Jim Brandes Multitest A Comparison of Probe Solutions for an RF WLCSP Product James Migliaccio RF Micro Devices This Paper Belgacem Haba, Ph.D. Invensas Wafer-Level Burn-in Decision Factors Steve Steps Aehr Test Systems COPYRIGHT NOTICE The paper(s) in this publication comprise the Proceedings of the 2013 BiTS Workshop. The content reflects the opinion of the authors and their respective companies. They are reproduced here as they were presented at the 2013 BiTS Workshop. This version of the papers may differ from the version that was distributed in hardcopy & softcopy form at the 2013 BiTS Workshop. The inclusion of the papers in this publication does not constitute an endorsement by BiTS Workshop, LLC or the workshop s sponsors. There is NO copyright protection claimed on the presentation content by BiTS Workshop, LLC. (Occasionally a Tutorial and/or TechTalk may be copyrighted by the author). However, each presentation is the work of the authors and their respective companies: as such, it is strongly encouraged that any use reflect proper acknowledgement to the appropriate source. Any questions regarding the use of any materials presented should be directed to the author(s) or their companies. The BiTS logo and Burn-in & Test Strategies Workshop are trademarks of BiTS Workshop, LLC. All rights reserved. BiTS Workshop 2013 Archive
Session 6 Bridging Between 3D and 3D TSV Stacking Technologies Belgacem Haba, Ph.D. Invensas Conference Ready 2/11/2013 2013 BiTS Workshop March 3-6, 2013 Content Introduction DRAM Application Package on Package Application Summary 2 1
Session 6 Packaging evolution Die Stacking BGA TSV die stacking Package Stacking 3 Packaging evolution Die Stacking BGA TSV die stacking Package Stacking 4 2
Session 6 Content Introduction DRAM Application Package on Package Application Summary 5 DRAM Challenge: 2012-2017 DRAM Growth: M Units Source: Gartner, DeDios 6 3
Session 6 Denser: Ultra-Book Memory Form-Factor SSD FAN DRAM Design Challenge: Shrink the Board to Increase Battery Size & Life DRAM 7 Dual Die DRAM Package Structures Opposing-Face DDP Face-Up DDP with FOW Face-Up DDP with RDL Dual Face Down (DFD) 8 4
Session 6 DFD Photos 11.5 mm square 9 RDL QDP Structures: Spacer and Stairstep ( QDP: Quad Die Package ) RDL QDP (with spacers) Stair step QDP (no spacers) 10 5
Session 6 New Quad Die DRAM Package QFD 17 x 17 mm x 1mm tall Quad Face Down Single-pass Wirebond Four DRAM die two spacer die No RDL, Short bondwires Thin construction: Two die tall No topside wirebonds Improved thermal properties 11 Form factor Standard SO-DIMM (2,640mm2) vs. DIMM-in-a-PACKAGE (394mm2) 12 6
Session 6 Faster: Monolithic Performance from All Die Bottom IC DQ Symmetric IC Performance 2133 MT/s (95C) DQ Top IC DQS/ DQS# Measured at 95C on production ATE (Advantest 5501) DQS/ DQS# Bottom die 2133MT/s (15 unit sample size) Top die 2133MT/s (15 unit sample size) 13 Significant Speed Bin Yield Gain Speed Bin Yield 1000 Unit Sample: Same Wafer Lot DDP = Dual Die Package DDP Control Structure 67% Delta 1333+ Dual Face Down Structure Dual Face Down: Bottom Die Dual Face Down: Top Die DDP Control 2133 67% gain in 2133MT/s speed bin yield vs. control DDP using same wafer lot. 14 7
Session 6 In-System Test Setup Agilent Infinium DSA91204A 12GHz EVGA SR-2 Classified motherboard Intel Xeon X5650 processor Digital Signal Analyzer For full-speed in-system probing of DQ, DQS 15 Functional At-Speed Testing DFD Module Tested Component Nominal Speed Die density & # die/module One DIMM/Channel (4 DQ loads/channel) Two DIMMs/Channel (8 DQ loads/channel) Invensas 8Gbyte Quadrank RDIMM w/dfd (1Gbit (x4)) die) 1333 MHz Components 1Gbit (x4 org x 2/package) (72 memory die/module) >1600MT/s 1600MT/s Market 8Gbyte Quadrank RDIMM with single die packages (2Gb: single x8 die) 1333 MHz components 2Gbit (x8 org x 1/package) (36 memory die/module) ~1600MT/s 800MT/s (barely operates) Market 16Gbyte Quadrank RDIMM with DDP packaging (2Gb: x4 die, DDPs) 1333 MHz components 2Gbit (x4 org x 2/package) (72 memory die/module) ~1600 MT/s 800MT/s (barely operates) 16 8
Session 6 DFD Reads DIMM1&2, 2DPC, 1600MT/s DQS (DIMM0) DQ1 (DIMM0) DQS (DIMM1) DQ1 (DIMM1) DIMM0 being read DIMM1 being read 17 DFD Writes at 1600MT/s and 2DPC DQS (DIMM0) DQ1 (DIMM0) DQS (DIMM1) DQ1 (DIMM1) 18 9
Session 6 Content Introduction DRAM Application Package on Package Application Summary 19 Processor-Memory Architecture The interconnect determines the computing performance and power usage 20 10
Session 6 Existing Processor-Memory Stacking Solutions PoP PiP TMV TSV The total market size for Package-on-Package stack was about 800M in 2010 Except for TSV, the stack packaging infrastructure is well established 21 Wide IO Roadmap 2010 2011 Mobile DRAM LPDDR 2012 LPDDR2 2013 2014 LPDDR3 Emerging 2015 2016? Wide IO Packaging PoP PoP PoP PoP TSV Mobile processor to memory interconnect 168 168 240 240 1250 Clock Speed (MHz) 400 533 800 200 Power 2X 1X 0.8X 0.5X Dual Dual Quad+ 8.5 12.8 >12.8 # of Channels Bandwidth (GBps) Single Single 1.6 4.2 Wide IO is approximately an order of magnitude increase in IO compared to current memory interface One method of implementing wide IO is using memory with TSV, which is not expected within the next few years 22 11
Session 6 Wide IO Roadmap 2010 2011 2012 2013 2014 2015 2016? Mobile DRAM LPDDR LPDDR2 LPDDR3 Emerging Wide IO Wide IO Packaging PoP PoP PoP PoP BVA PoP TSV Mobile processor to memory interconnect Clock Speed (MHz) 168 168 240 240 400 533 800 Power 2X 1X 0.8X IO ranging from 200 to 1000+ High IO offers high bandwidth at low speed Enables intermediate power reductions 1250 The goal of BVA PoP is to offer TSV capabilities for PoP applications utilizing conventional PoP infrastructure and materials 200 0.5X # of Channels Single Single Dual Dual Quad+ Quad+ Bandwidth (GBps) 1.6 4.2 8.5 12.8 >12.8 >12.8 23 BVA Stand-off issue eliminated: Wire-bond based memory-logic interconnect 1000+ wide IO: 0.2 mm pitch easily possible High performance at low-cost: Conventional PoP materials and processes 24 12
Session 6 BVA PoP Scalability Pitch (mm) No. of IO rows 0.50 0.40 0.30 0.25 0.20 2 200 248 336 408 512 3 4 5 6 288 360 492 640 600 784 960 756 992 1220 1440 Assigning the same amount of area for IO as that of the current 0.5 mm pitch PoP, BVA with 0.2 mm pitch can offer up to 1440 IO 25 Overmold and Wire Exposure Process Average wire protrusion height Protrusion Height (um) Top view of Overmolded bottom package 26 13
Session 6 Test Vehicle Assembly: PoP Stack Top surface of bottom package Fully Assembled BVA PoP Package The top surface of the of the bottom package has bond wires projecting outwards by about 0.1 mm. The two packages were joined using conventional PoP SMT approach 27 Differential impedance Different technology implementation for target impedance Dbva (um) D 2 differential SIGs GNDs all around 50 50 50 50 50 Differential S (um) Impedance (ohm) 240 125 280 135 320 144 360 151 400 158 S S 28 14
Session 6 Differential return loss and insertion loss Return loss Dbva (um) 240 um S (um) 50 240 50 400 400 um Insertion loss 240 um 400 um BVA itself is not too big an issue for RL and IL assumed 550 um BVA height 29 Summary Demonstrated two technologies that can bridge the gap before the TSV technology arrives The two technologies offer platforms for higher bandwidth Platforms that uses available assembly and available testing infrastructures 30 15