A High-Level ‘How To’ Guide For Effective Chip-Package Thermal Co-Design

Seven important steps you need to build into your current and future flows deal with heat.

popularity

By John Parry and Byron Blackmore
Concurrent design of a chip and its packaging environment is becoming more important than ever for several reasons. Designing a large, high power die, e.g. a System-on-Chip (SoC), without considering how to get the heat out is likely to lead to problems later on, resulting in a sub-optimal packaging solution from cost, size, weight and performance perspectives. Historically IC design has considered the die temperature to be uniform. This is no longer a valid assumption in many cases. Heating due to current leakage, which is temperature dependent, is making power dissipation less uniform, and the use of thinner die, now well under 50µm, has reduced the heat spreading capability of the die itself. Both of these effects contribute to greater on-die temperature variation.

Chip-package co-design becomes essential when designing stacked Three-Dimensional Integrated Circuits (3D-ICs). The dies cannot be designed independently due to their electrical and thermal interaction. Through Silicon Vias (TSVs) that act as inter-die interconnections can help get heat out of the die stack, although their primary thermal role is in reducing hot spots, so their placement relative to high power regions on the die can have a marked effect on the overall thermal performance. Conductive heat transfer is a highly 3D phenomenon so the package temperature distribution affects the temperature distribution on the die.

IC and package designers can go a long way toward optimizing chip-package co-design by following some basic procedures, which we’ll layout as seven major steps in the remainder of this article.

1: Always Start with the Package
To get the correct temperature distribution within the die, it’s essential to include the package construction in the thermal model, mounted on a typical PCB and where appropriate with a representation of a heatsink solution, so that the effect of heat spreading in the board and into the heatsink are accounted for in the predicted package temperature distribution. For this a full computational fluid dynamics (CFD) simulation is required to correctly predict the way the package interacts thermally with its environment, and hence predict the correct temperature distribution within the package itself.

In early design, and before the detailed IC design starts, there is the greatest scope for optimizing the chip-package architecture. At this stage the number of die, and the intended size and budgeted power for each die will be known. This can be used to create a 3D thermal conduction model of the candidate package(s) that can be used to explore the package design space. The model could also include convection and radiation if the package style includes an internal cavity. It also provides the thermal environment that will allow temperature data to be back-annotated to the IC design flow.

Figure 1. Detailed Thermal Model of a BGA Package with Multiple Die.

Figure 1. Detailed Thermal Model of a BGA Package with Multiple Die.

2: Explore the Package Design Space Before the IC Design Starts

The thermal model of the candidate package(s) can be used to investigate the influence on the thermal performance of different die arrangements, package size and packaging materials, for example the amount of copper in the substrate for a Ball Grid Array (BGA) package.

At this stage there is a high degree of freedom, and hence opportunity, to explore different package options, and the design of each. Based on an initial estimate for the size of the die, design parameters that can be investigated to determine their influence on the die temperature rise and variation for a given package style may include:

  • Influence of the number and possible layout of TSVs
  • Influence of the size, shape, and material choices for an interposer
  • Effect of interfacial resistances (glue layers)
  • Pyramid vs. overhang stack arrangements (if wire bonded)
  • Cooling solutions internal to the package, e.g. die edge cooling, internal heat slugs, etc.
  • Influence of external cooling solutions, e.g. solder pads, underfill options etc.

Features that have the greatest effect on die temperature rise and variation self-indicate the need to be modelled in greater detail, and optimized for thermal performance.

3: Include Temperature Dependent Thermal Properties
This is automatic for silicon and other materials included with thermal design tools, such as Mentor’s FloTHERM® material library. Temperature ranges across the die are likely to be too high to assume a single thermal conductivity value, so temperature-dependent thermal conductivities are necessary to accurately predicting die hot spot values. Note that for transient calculations it is essential to include the material density and specific heat capacity. This happens automatically in FloTHERM when a material is attached to an object.

4: Refine the Die Surface Treatment
Include a 3D representation of the active layers of the die (metallization and polysilicon) with an isotropic block ~0.5 to 1.0 µm thick.

Silicon dioxide (SiO2) and silicate glass, typically used as dielectric materials to separate metal wires on the active surface of the die have thermal conductivities of the order of 1 W/mK, around two orders of magnitude less than the metal they isolate, historically being aluminium or copper. Wires on different levels run in different directions, so the material behaviour is locally orthotropic. However, the high level of interconnection between the levels, combined with metal running in different directions, causes the heat to smear. Hence for early design activities outside the main IC design flow the bulk behaviour can be approximated with an isotropic material with all the active surface layers captured within the thickness of one mesh cell in the package-level model.

The IC process and design technology files contain information about metal width and spacing as well as the preferential routing directions. This can be used to calculate the overall thickness and an isotropic averaged material property for this thermally-active layer.

5: Back-Annotate Temperature Information Before Floorplanning
Up to this stage heat should be distributed uniformly over the die. This will not be the case in practice, and the model should be refined to remove this assumption as soon as more detailed information is available from the IC design team. The benefit of using this assumption at the outset is that it gives an indication of the inter-die temperature variation that arises from the package’s limited ability to hold the bulk die temperature uniform.

Providing the IC design team with information about the average die temperature and temperature variation for each die before the IC design process starts can greatly help floorplanning [Ref. 1], which is critical to the quality of the design, as decisions made during floorplanning can either alleviate or exacerbate this temperature variation.

6: Use Power Budgets During Floorplanning
Once floorplanning starts, get a high-level power map from the IC design team and import that into the thermal model of the package. For example, FloTHERM has a Die SmartPart that allows powers to be read in as a CSV file so this can be done automatically and the results quickly fed back as the simulation model will often only take a matter of minutes to run, indicating where TSVs can be introduced to improve the thermal performance, or where design changes are needed. For example, it may be important to ensure that two or more different functional blocks operate at very similar temperatures to eliminate timing issues.

For logic-on-logic 3D-ICs this should be accounted for when partitioning the design amongst the various dies, and during inter-die and intra-die floorplanning, requiring power map information for each die. At this stage opportunities exist to move the functional blocks in both x- and y-directions (xy expansion) keeping their relative positions the same but adjusting the gaps (white space) between them into which TSVs can be inserted to examine their impact on die hot spots. Knowing the TSV size and pitch, which scale with die thickness, blocks of higher through-plane thermal conductivity can be superimposed over the die thickness in these white space regions in to locally override the properties of silicon.

In early floorplanning extremes can be investigated to inform the IC design team of the extent to which hot spots can be controlled using TSVs, bounding the problem. Optimizing both functional block and TSV layouts during floorplanning needs to be done as part of the IC design flow.

Figure 2. Die Temperature Distribution Due To Non-Uniform Power Map.

Figure 2. Die Temperature Distribution Due To Non-Uniform Power Map.

7: Make the IC Design Flow Temperature Aware
As floorplanning progresses, the thermal design effort needs to focus on the detail of the thermal interaction between die as the design is further elaborated. The power map for the die becomes much more detailed, and in the case of a 3D-IC, the number and location of TSVs need to be defined as part of the electrical design.

Figure 3. Opportunity for Parallel Package & IC Physical Design Flows [Ref. 2].

Figure 3. Opportunity for Parallel Package & IC Physical Design Flows [Ref. 2].

This work is best done within the IC design flow itself, and to this end FloTHERM has been embedded into Mentor’s Calibre® suite, enabling FloTHERM results to be viewed in the Calibre DESIGNrev™ and RVE™ physical verification results viewing environments [Ref. 3]. This provides an easy-to-use, fast and accurate flow to do thermal simulation on dies and interposers of 3DIC, enabled by automatic gridding that is built into this solution that uses a localized grid in critical model areas such as the dies.

This solution takes die power map files that can be generated by power analysis tools, and creates thermal maps that can be used for thermal design and for checking against thermal constraints. For example, thermal results can be displayed in a histogram in Calibre RVE, and the thermal hotspots can be highlighted in the design in Calibre DESIGNrev. For transient analysis, temperature vs. time graphs can be displayed using Mentor’s EZwave™ graphical waveform environment.

More accurate thermal models of dies that take into account the metallization, such as interconnects and TSV in the dies, can be created and used in thermal simulation. A 3D-IC thermal model can be created to allow the 3D-IC package to be imported into a larger system, for further thermal simulation at the system level.

Acknowledgements:
Lee Wang, Technical Marketing Engineer, Calibre Marketing, Mentor Graphics
Marko Chew, Technical Marketing Engineer, Calibre Marketing, Mentor Graphics

References:

  1. Floorplan (microelectronics). http://en.wikipedia.org/wiki/Floorplanning#Floorplanning
  2. Physical design (electronics). http://en.wikipedia.org/wiki/Physical_design_(electronics)
  3. Mentor Graphics Provides Design, Verification, Thermal and Test Solutions for TSMC’s CoWoS Reference Flow, Mentor Graphics Press Release, October 15, 2012. http://www.mentor.com/company/news/mentor-tsmc-cowos-reference-flow

Authors

John_Parry_UK

Dr. John Parry, CEng is Electronics Industry Manager with Mentor Graphics Mechanical Analysis Division. A chemical engineer by training, John serves on the JEDEC JC15 Thermal Standards Committee and on various conference committees, and was General Chair of the SEMI-THERM 21 conference.

Byron Blackmore

Byron Blackmore is the Product Manager for the FloTHERM product at Mentor Graphics. Byron has been with the Flomerics Mechanical Analysis Division for nine years, and has previously acted as the Electronics Cooling Engineering Supervisor for North America. He has a BS in Mechanical Engineering from the Technical University of Nova Scotia and an MS in Mechanical Engineering with a concentration in fluid dynamics and heat transfer from the University of Alberta.