3D By Design: Are Path Finding Tools the Missing Link?

3D By Design: Are Path Finding Tools the Missing Link?

TechSearch International’s recently published 3D IC Gap Analysis report has a section for a new class of tools called Path Finding, which are separated from the classical EDA tools. The classical EDA tools discuss implementation and verification for IC, package, and PCB. Path Finding tools are used well before the Implementation/Verification cycles begin and should help reduce the number of cycles, resources, and time required before release to manufacturing. One ‘all-in-one’ Path Finding tool does not exist and current Path Finding tools focus in specific areas. Path Finding solutions exist for: RTL/Floorplanning, SI/PI electrical analysis, thermal analysis and mechanical. A cost modeling of the various Path Finding test cases is also required to ensure that cost targets are achieved for the end product.

Design Flows and Interfaces Resolved…..
A technical cost modeling paper written by SavanSys Solutions LLC titled: “Cost Trade-off Analysis of Pop versus 3D TSV”, discusses EDA co-design barriers where each EDA vendor’s tools do not easily interface between each other. SavanSys was not talking about tools within an IC or package or PCB environment; but rather when an IDM uses one EDA supplier’s tools for IC design, another EDA supplier’s tools for package and possibly a third EDA supplier’s tools for PCB design. Trying to span the chasm between these design areas is difficult at this point and requires lots of manual work to enable co-design/co-verification. Since Path Finding tools are used BEFORE the actual implementation, importing/exporting of design data is not required. So the IC/Packaging/PCB database and interface issues do not exist. In addition, this implementation chasm is AFTER significant resources and time have been invested implementing a design: a KEY reason why Path Finding tools can be invaluable early in the planning phase.

….But this causes other issues for Path Finding tools
But solving these database/interface issues places a burden on Path Finding tools: they need to work with many different materials (examples: glass/silicon interposers, balls/pillars of various sizes and materials, wire bonds of various materials, diameters and lengths, etc). Each material has its own physical and electrical properties that need to be accurately and objectively analyzed to help find that optimum solution. In addition, the complexity and size of these structures require CPU and memory efficient algorithms that allow larger structures to be analyzed: simulating a few TSVs is insufficient for accurate analysis of a structure.

Promising future?
The EDA industry continues to improve the links between implementation and verification. Path Finding tools have the potential to resolve issues faced early in product development planning. I believe that Path Finding tools are the missing link between Planning and Implementation. Is P=IV a coincidence? Path Finding “Power” applied to Implementation and Verification? Time will tell as product development explores alternatives to homogeneous silicon and how efficient and effective they are exploring. ~ B. Martin

  • Dr. Dev Gupta

    For years the EDA companies have been feeding off the fundamental development ( “pathfinding” ) of physical hardware and their characterization at IDMs and then turning them into rule based design software for Fabless Co.s. It has been a sort of “free lunch”. Well the “no device R&D” model of Fabless Co.s have made them successful at the expense of most of the IDMs. So a lot of the effort for TSV based 3-d die stacking has been happening outside the IDMs, but in a technically fragmented way. There is a lot of “local” optimization but not much “integrated / global” optimization, which is the principal reason behind the delays. Given the current competitive scenario the 3 large IDMs that have been engaged in TSV based 3-d stacking may not feel generous enough to let their IP once again percolate to the Fabless Co.s through these EDAs and Foundries.

    • Dr. Gupta,

      My first 15 years in the industry was on the IDM side and I saw the growth and maturity of EDA as a partner with IDMs. My last ~20 years has been on the EDA/Service side. I think I have a perspective from both sides and have seen the pluses and minuses.

      If you roll back 20 years (?) ago, you will see the growth of standalone silicon foundries. Long ago, many companies had their own fabs. Companies were not considered ‘big boys’ unless they owned their own fab. But the Fabless market quickly grew based on using 3rd party Foundries (such as TSMC). Yes, I saw these struggles at Mostek and VLSI Technology. Each had their own issues and VLSI started co-development relationships to ease the process investment costs.

      There is a long history of EDA productizing algorithms into useful tools that could serve the masses. Vertically integrated IDM companies learned in the 1970/80’s that they spent inordinate amounts of resources ($s, people) trying to manage, develop and support tools that could aid their engineering teams. IDMs learned that most EDA tool development would better supported by 3rd party vendors solely focused on SW tools. Several large IDMs have sold their tool development (patents, algorithms, personnel, etc ) to EDA companies. As IDMs reduce their internal EDA SW development efforts, they refocused their efforts on product development (the needs of their end customers). Rather than go through the annual budget process on what tools to develop and support, it was easier to benchmark and purchase the best tool provided by several vendors. Competitive pressures in the EDA force EDA vendors to continually improve and release new products that their customers evaluate for purchase. If this was not successful, you would see many more IDMs shrink their EDA external budgets and re-establish internal development teams. Just the opposite has happened. The EDA market continues to growth. I saw this at Mostek and VLSI Technology, first hand. .

      Recently, you are seeing a similar phenomenon concerning Silicon IP. Many companies that used to develop their own IP are now looking externally to purchase IP. Same type of reasons: why divert limited, precious resources on something that can be purchased; focus resources on development to differentiate their product in the market; cost, quality and support can often be better served by external 3rd parties. Again, some IDMs have made arrangements with IP vendors to take over their product lines. Recently, AMD announced an arrangement with Synopsys. In the past 5-10 years, we have seen explosive 3rd party IP revenue growth that is double digits and surpassing EDA tools’ growth. Again, the market is deciding which is the most efficient and effective method for IP. In one of my companies/jobs, I was approached by several large companies to purchase and take over their IP divisions and these were multi $B companies. They wanted to focus on areas they could differentiate, not produce, maintain, port and support IP. IP was a ‘cost to them’, not directly a revenue generator.

      So you have the entire system supporting dis-aggregation: allowing each participant to focus where they can maximize their value in the system. I have not heard many IDMs or Foundries or EDA or IP providers or even end customers raise significant and prolonged arguments to go back how it was 30 years ago. The new ecosystem allows many new companies to be founded without requiring a huge investment to cover all the development needs. Rather than building their own fabs, creating their own EDA tools, creating their own IP libraries, etc, they can quickly enter the market focusing on their strengths.

      Each company will determine what they believe to be their best strategies for success. Outsiders may or may not agree with their, some insiders may not agree! The true test will be several years later based upon their strategy execution and how the market reacts (P&L results).

      How the entire eco-system will look in a few years, I have no idea. But I believe the market will perform ‘self’ corrections. If processes or tools or IP or other products/services do not meet required customers’ schedules, quality or support expectations, other suppliers will be given a chance. Once no external supplier can fulfill the need, then the owner will consider taking on the ownership internally.

      Just my two cents.