Dialog on industry challenges and university research activities among technologists from Participating Companies, Students and Faculty

Transcription

1 IMPACT Internal Document for IMPACT Participants Only Summary IMPACT Roundtable Lithography + DfM Dialog on industry challenges and university research activities among technologists from Participating Companies, Students and Faculty Held at SanDisk, Milpitas September 24, 2008 For further dialog contact the IMPACT faculty: Puneet Gupta, Andrew Kahng, Andy Neureuther, Kameshwar Poolla, Costas Spanos University of California Berkeley San Diego Los Angeles 1

2 IMPACT UC Discovery Industry Team Thanks! University IMPACT Lithography/DfM of California Berkeley Roundtable San 2 Diego Los

3 Roundtable Approach Objective: guide research in the IMPACT program understand litho & DfM challenges in industry understand research capabilities in universities what are issues that are relevant and timely to industry and academic in nature Format: 2 hours of presentations (8 industry, 5 faculty) Summarized by speaker suggestions 2 hours of dialog 45 participants (30 ind.15 univ.) Summarized by topics (DfM, Variability, Physics) IMPACT Lithography/DfM Roundtable 3

4 Speakers Industry Lars Liebmann, IBM, ASIC Chandu Gorla, Flash Memory, San Disk Vivek Singh, Microprocessors, Intel Zhenhai Zhu, Double Patterning, Cadence Huixiong Dai, thin-films for pattern transfer, AMAT Mamoru Miyawaki, Exposure tools, Canon Bob Socha, Exposure tools, ASML Stan Stokowski, Inspection, KLA-T University of California Andrew Kahng, UCSD DfM tools Puneet Gupta, UCLA DfM tools Kameshwar Poolla, UCB Algorithms Costas Spanos, UCB variability and assignment of causes Andy Neureuther, UCB fast-cad, electrical test, EM IMPACT Lithography/DfM Roundtable 4

5 Main Points DfM is all about knowing how much performance to leave on the table to make the IC manufacturable. DfM (or computational Lith., etc.) is expected to provide a full generation now that lithography is no longer scaling. DfM is time-changing as DfM results acted on in the fab improve the process but design requires models extrapolated to the future improvement at tapeout date. DfM helps manage complexity Support playing what-if games at macro or product level Need to make tolerance consequences and budgets visible Scaling layouts to the next generation no longer works - requires design Double patterning technology is nearly here for the 22nm node (logic and flash) and 32nm node (memory) as a necessary reality. The choice of DP method depends on product and tolerances including overlay DP methods require rethinking layout concepts and metal layers in design Thick mask EM effects must be mitigated Need better models of all physical processes including 3D effects Simulated assisted metrology is required IMPACT Lithography/DfM Roundtable 5

6 Mechanical Models for Long Range Stress Effects IMPACT Lithography/DfM Roundtable 6 Lars Liebmann, IBM Challenge: DfM needs to reflect the improvement of process with time. Once actionable, modeled, systematic sources of variation are found through statistical methods (of DfM) they may be corrected in either Fab or Design. What the designer wants is not a current but a future model of process maturity at the tapeout date for their product. Research Suggestions: (Some of these are detailed in later slides) Process Variation Analysis, Decomposition, and Correlation that would facilitate more focused process or layout optimization More Rigorous Approximation to Mask EMF Effects including high- NA and parallelized codes Extension of fast first principle predictive modeling solutions to etch, CMP, implant, anneal Simulation Assisted Metrology for understanding inspection and improving extraction of meaningful metrics

7 Chandu Gorla, SanDisk Challenge: In Flash memory generations are only 12 months apart Immersion is in production, spacer pattern doubling is nearly here, and EUV and Imprint are just around the corner Suggestions: Materials for double patterning: freeze or multiple exposure; single layer double exposure Overlay Odd and even pattern dependence: understanding and modeling the interaction between litho and etch DfM needs to be flexible to include novel integration schemes and their tolerances EUV and Imprint need study IMPACT Lithography/DfM Roundtable 7

8 Vivek Singh, Intel Comment: The following suggestions emphasize modeling and algorithms. Topics that are theoretical are easy to transfer quickly to industry. Suggestions: Recovering the lost 3 rd dimension (resist thickness and mask edge effects). Litho effects lost in the popularity of OPC with only 2D effects Mask scattering to include thick-mask effects at advanced nodes Inverse Lithography that is fast and not too fragmented Optimization algorithms that are fast for sizeable numbers of discrete variables Litho/Etch coupled models with innovation in fast formulations for the complex physics of etch. IMPACT Lithography/DfM Roundtable 8

9 Zenhai Zhu, Cadence Challenge: Splitting a layout for double patterning is not just a tapeout problem Scaling often requires going back into the design for contacts, redundant vias, etc Restricted Design Rules are often ridiculously restricting Suggestions: Automated netlist to physical layout Single-contact process (Reliable litho, deposition, electroplating of small holes) IMPACT Lithography/DfM Roundtable 9

10 Huixiong Dai, Applied Challenge: Double patterning is the only feasible solution for 32nm HP and beyond before Meeting NAND Spec (+/-10%) for 32nm and 29nm Half-Pitch 26nm and 22nm is limited by resist LWR Suggestions: Sidewall Spacer Double Patterning with non-gridded Design requires Design Evolution for Logic Highly regular gridded designs are promising for 44 nm pitch but a greater patterning concern is the isolated slots Resist LWR is an industry wide challenge and it interplays with the core and space tolerances IMPACT Lithography/DfM Roundtable 10

11 Mamoru Miyawaki, Canon Challenge: Future nodes such as 22 nm require very careful examination of techniques and tolerance budgets. Suggestions: Verification methodology to reproduce the actual experimental data by DfM software. (How many parameters? Accuracy for each?) DFM provide a system for understand every aspect of the whole process and error budgets Methods for monitoring very low lens aberrations and assuring that aberrations have virtually zero effect on lithography. IMPACT Lithography/DfM Roundtable 11

12 Bob Socha, ASML Challenge: DPT for the 22nm node (logic) and 32nm node (memory) is a necessary reality Suggestions: Overlay is critical in double patterning and in vertical scaling (3D layout) Illumination source and Mask Optimization needs to interact with designers, fab, and scanner vendor How can this interaction knowledge be included in the design library and in the place and route CAD tools IMPACT Lithography/DfM Roundtable 12

13 Stan Stokowski, K-T Challenge: It is not just a matter of seeing the variation but rather how this information can be used to optimize all of the settings Suggestions: Understand IC process control on a broader basis and determine what to measure and inspect for controlling the process Lithography simulation of mask 3D effects at the wafer Identify issues in EUV in advance Identify issues in Imprint in advance IMPACT Lithography/DfM Roundtable 13

14 DFM Wishlist Discussion 1 (Gupta, Kahng) #1. Accurate, quick map from design rules to product quality (area, power, frequency, yield, (COO)) (Rob Aitken, ARM) need an infrastructure within which to play such what-if games at macro or product level (Lars Liebmann) Framed by SPICE/ITRS parameters, or TCAD, or process knobs? physical parameters (Lg) preferred to electrical (Vt) for informing process (Vivek Singh, Lars Liebmann) #2. New ways for design to ease required process control capability (Zhilong Rao, AMAT) #3. Design flow integration of systematic variation models Orthogonality of designs to systematic variations (related to above #1, #2) + Metrology to support But, aren t big systematic issues solved in the fab? (Lars Liebmann) random variations increasing in significance (JL de Jong, Xilinx) Wafer-level systematic variation is less well-addressed than die/field-level where OPC, PPC have had success (Costas Spanos) #4. Research on DPL bimodal problem Xilinx: already study mismatch; assume have to extend to DPL context ASML / Socha: feel this is valuable work; make sure to cover overlay as well as CD DPL flavor: bimodal distribution of space, vs. bimodal distribution of width IBM (Lars): bimodal problem is unacceptable, and hence uninterested in this process if such a problem exists(!) assume that this is controlled Interconnects also of interest (Lars); ongoing at UCLA, UT, N.B.: even if distributions perfectly matched, still lose spatial correlation Solving a single-layer formulation may be irrelevant! (Apo, Cadence) University solvers: transferability of code is of interest (IBM, ASML,...) IMPACT Lithography/DfM Roundtable 14

15 DFM Wish list Discussion 2 (Gupta, Kahng) #5. Layout automation for fixed pitch / fixed orientation (JL de Jong, Xilinx) May be beyond capabilities of university research? (Lars) But, prototypes of new ideas possible? (Andy) IMPACT can at least help industry make a decision Cf. Rob Aitken wishlist #1 #6. Miyawaki-san (Canon) comment: How do I determine my budget out of a DFM tool? Issue today: electrical budgets not necessarily well-aligned to process budgets Cf. LEE rule / taper shape / leakage-area correlation of metrics UCLA/UCSD Can you even establish yield impact of a litho hotspot? (Lars) Can you tell me the exact yield impact of double-cut vs. single-cut vias? (Lars) (Berkeley work (Spanos, Poolla) is responsive to these types of questions...) #7. (Huixiong Dai, AMAT) Have you looked at layout that is specifically directed at spacer double-patterning? UCLA is starting to look at this a bit (e.g., M1 patterning) maybe some insights within a few months General comments Time horizon? Tractability to students? Nature of application/specific research: artifactual, less generalizable or impactful than a new framework or theory? (Vivek) Ultimately, DFM tools must become more integrated into overall process planning and budgeting (Andy) IMPACT Lithography/DfM Roundtable 15

16 Metrology and Algorithms Wish list Discussion 3 (Poolla and Spanos) #1 Uncertainty in Circuit Simulation Is there a need to go beyond the 4 corner model of spread in performance? (Yes) Is there a need to propagate uncertainty through models that are nonlinear (max)? How are results of uncertainty characterization reinserted into design flow? Does not matter as both design and process have limitations. This is parametric loss Understanding robustness to estimation errors. #2 Accuracy of Circuit modeling Are all the models under zero variability conditions that go into tool sufficiently accurate to be predictive of circuit performance. SPICE 15% to 20% (but derivative is more accurate.) Voltage and temp may dominate corner speed estimates more than process Designers might choose devices with low variability (fixed pitch price) this needs to be included in Circuit simulation options in SPICE) Tool to help make a decision before a layout exists. Interaction between design models and layout and need to include and quantify other factors such as temp. Corners assume convex hull but possibility that worst situation may be in the middle (temp). IMPACT Lithography/DfM Roundtable 16

17 Metrology and Algorithms Wish list Discussion 4 (Poolla and Spanos) #3 What is DfM? DfM is all about how much you leave on the table to make it manufacturable and knowing how much that is. Looking for us to estimate the performance benefit of the best possible variation reduction. (How much faster would the circuit run?) Alternatively to modeling uncertainty DfM is about reducing variation and could use the 5 tools described. If wavelength gap had not increase would we have had a need for DfM? How much is Litho Today? #4 DfM Scope Does place and route use models? Real data driven used upstream in design? Need layout checked afterward. Monte Carlo is slow is there sampling ways to speed it up. #5 Roles for DfM At 90 nm processes are well characterized yet full process aware design might result to the improvement of a full generation. The next generation plans to get the realizable speed extrapolation from the previous generation but process discontinuities (metal or new foundry) require a new guess. This is a challenge for data driven modeling. A solution is to keep layout and recalibrate on 10% structures, group and see effects, and focus on possible causes at length scale. Apply machine learning to translate existing (SEMs) to changes in process IMPACT Lithography/DfM Roundtable 17

18 Litho Wish list Wish list Discussion 5 (Neureuther) #1 Sources of Variation: Methods for identifying variations due to illumination in addition to focus and dose Masks: More Rigorous Approximation to EMF Effects Overlay and propagation in 3D Methods for aberration measurements and assurance of zero contribution to patterning #2 Process Modeling Extension of Process Modeling {Litho 3D, Fast Image {ideas from other fields}, Etch, Anneal, CMP} Mechanical Models for Long Range Stress Effects on wafer distortion Simulation Assisted Metrology CD SEM is uncaligbrated compared to artifact (10 nm uncertainty) 3 rd Dimension is lost here as well #3 Future Process {equivalent of negative resist?} Look ahead to understand issues in EUV and IMPRINT 3D Integration Double patterning (reach through, CMP, resistance of contact both yield and circuit performance) Some work on stacking but not being pursued yet Fixed pitch small circuits study on how well work and how much leave on table (EDA Ind) Need automatic tool to relate to 3D configuration (EDA ind) IMPACT Lithography/DfM Roundtable 18

19 Liebmann: Process Variation Analysis, Decomposition, and Correlation Given sets of empirical data taken across multiple lots, wafers, chips, and field locations, collected after various process steps (e..g. resist, etch, anneal) by various means (e.g. sem, electrical) establish frequency, process, and layout correlations that would facilitate more focused process or layout optimization. The goal would be to develop an analysis and visualization software platform that combines layout information, process simulations over different length scales, and data analysis to pinpoint error sources. Detailed understanding of correlation distances may also improve incremental layout optimization (e.g. cell-level, macrolevel, chip-level). IMPACT Lithography/DfM Roundtable 19

20 Liebmann: More Rigorous Approximation to EMF Effects Need sufficiently accurate EMF calculations to incorporate into iterative optimizations. Accuracy of domain decomposition is being questioned at ultra-high NA and extreme illumination angles. Code should focus on extreme parallelization, potentially running on supercomputing systems such as CCNI. Also, need to make Tempest 7.0 available. IMPACT Lithography/DfM Roundtable 20

21 Liebmann: Extension of Process Modeling Build fast first principle modeling solutions for processes other than lithography (e.g. etch, cmp, implant, anneal). Models should be physical enough to allow accurate predictive modeling. IMPACT Lithography/DfM Roundtable 21

22 Liebmann: Simulation Assisted Metrology Build models to gain first principle understanding of metrology image collection and data extraction to improve accuracy against meaningful metrics (e.g. top-width, bottom-width, sidewall angle). IMPACT Lithography/DfM Roundtable 22

23 IMPACT Lithography/DfM Roundtable Thanks to All Speakers for preparing comments Technologists for your comments and interactions Students for doing the hard work! SanDisk for hosting and Starleigh Arce for local arrangements Changrui Yin for web site posting and reservations Reminder: IMPACT Workshop, Wed. October 29 th at the AMD Commons Building, Santa Clara IMPACT Lithography/DfM Roundtable 23