Thermal limits and benefits offered by different packaging options.
Packaging density, electrical performance and cost are the primary factors driving electronic package architectures for high-performance server markets. Considerations such as thermal performance and mechanical reliability are equally important but tend to be addressed later in the design cycle. Presented in this paper is a historical view of the packaging trends leading to the current multi-die package options. Particular attention must be placed on the thermal limitations and benefits offered by each design. Since multi-die packages have many junctions of interest, a method for characterizing the package with arbitrary power conditions is required. A superposition method, using a matrix approach, is presented that will enable the end-user to investigate the effects of power levels on junction temperatures. Experimental data measured on a 2.5D package were taken to demonstrate the matrix approach for predicting junction temperature based on an independent power map. The agreement between the matrix model and data generated by an independent power map is within 8%.
To read more, click here.
Interest in the open-source ISA marks a significant shift among chipmakers, but it will require continued industry support to be successful.
Slowdown due to impact on timing, and dependencies between power, thermal and timing that may not be caught by signoff tools.
Access to source code makes it attractive for custom applications, but gaps remain in the tool flow and in software.
SoC bandwidth, integration expand as data centers use more FPGAs for machine learning.
Problems won’t derail next-gen litho, but could limit use and affect schedules.
AI-enabled systems are being designed to process more data locally as device scaling benefits decline.
Access to source code makes it attractive for custom applications, but gaps remain in the tool flow and in software.
Growing volume of data and limited improvements in performance create new opportunities for approaches that never got off the ground.
What happens when companies are combined? The outcome often isn’t as good as the announcement.
As complexity and device sizes rise, so does the need for an on-chip network.
Experts at the Table, part 1: Are we looking to solve the wrong problems? Where does parallelization work best and why.
Synopsys Co-CEO Aart de Geus explains why systemic complexity will be the next big challenge, and what that means for EDA and the chip industry.
IP reuse has reduced the urgency for a higher level of abstraction in complex system-level design, but that still could change.
 |
 |
 |
 |
 |
 |
 |
 |
 |
|
Leave a Reply