Optical HBM Disaggregation for ASIC Space and Power Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-performance computing requires multiple high-bandwidth memory (HBM) dies, but integration in ASIC packages is limited by space, power, and thermal constraints, leading to costly and wasteful replacement of entire packages due to faulty HBM dies.
Innovation Solution
Disaggregating HBM dies from ASIC packages by optically connecting them through HBM optics module packages, which include HBM dies, chiplets, and optical chiplets, allowing for flexible placement and reduced repair costs by replacing only faulty components.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If HBM dies are integrated within ASIC packages, then bandwidth and capacity are improved, but space constraints, power constraints, and thermal constraints of the ASIC package worsen
Solution Approach 1:
The system divides the integrated memory-processor package into separate modules: ASIC packages and HBM optics module packages. This segmentation allows each component to be optimized independently and connected via optical links, resolving the space constraints while maintaining high bandwidth through disaggregated memory architecture.
Solution Approach 2:
Optical fiber links serve as an intermediary between ASIC packages and HBM dies, enabling high-speed data transmission without the physical constraints of direct integration. This intermediary solution allows memory and processing units to be separated spatially while maintaining performance.
2Productivity
If additional HBM dies are integrated in ASIC packages, then bandwidth is improved, but power constraints and thermal constraints of the ASIC package worsen
Solution Approach 1:
By segmenting the system into separate ASIC and HBM packages connected optically, power consumption is distributed across multiple independent modules rather than concentrated in a single integrated package, allowing better thermal management and power distribution.
Solution Approach 2:
The solution transitions from vertical integration (stacking HBM dies within the ASIC package) to horizontal distribution (separate packages connected via optical links), effectively moving the architecture to another dimension where power and thermal constraints are managed differently.
3Productivity
If HBM dies are integrated within ASIC packages, then performance is improved, but reliability worsens due to defect propagation
Solution Approach 1:
The system architecture segments memory functionality into independent HBM optics module packages that can be individually replaced. This segmentation isolates defects to specific modules rather than propagating them across the entire system, improving reliability through modular fault isolation.
Solution Approach 2:
The system changes the operational parameters by using optical communication instead of direct electrical integration, creating electrical isolation between ASIC and HBM components. This parameter change prevents defect propagation through the interface while maintaining high-performance data transfer.
4Reliability
If entire ASIC package is replaced due to faulty HBM die, then system reliability is restored, but cost and waste increase
Solution Approach 1:
The system is segmented into replaceable modular units where only the faulty HBM optics module package needs replacement rather than the entire ASIC package. This segmentation enables selective component replacement, reducing waste and cost while restoring system reliability.
Solution Approach 2:
The HBM memory functionality is extracted as a separate, independently replaceable module from the ASIC package. This extraction allows the memory component to be removed and replaced without affecting the processor or other system components, minimizing loss and waste.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables increased flexibility in HBM die placement, reduces repair costs, and maintains performance by extending the channel reach of connections, allowing more HBM dies to be connected to ASIC packages without enlarging the package, and reducing downtime and costs associated with faulty components.
Implementation Method 1
The HBM optics module package may be connected to an ASIC package via one or more optical links. The optical chiplet may be configured to connect the HBM optics module to one or more optical fibers that form an optical link with one or more other components of the ASIC package.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The technology generally relates to disaggregating memory from an application specific integrated circuit ("ASIC") package. For example, a high-bandwidth memory ("HBM") optics module package may be connected to an ASIC package via one or more optical links. The HBM optics module package may include HBM dies(s), HBM chiplet(s) and an optical chiplet. The optical chiplet may be configured to connect the HBM optics module to one or more optical fibers that form an optical link with one or more other components of the ASIC package. By including an optical chiplet in the HBM optics module package, the HBM optics module package may be disaggregated from an ASIC package.