Ferroelectric Memory Chiplet Layout for AI Bandwidth and Thermal Limits

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing AI processing systems face limitations in I/O bandwidth and thermal management due to the stacking of dynamic random-access memory (DRAM) on top of a compute die, leading to periphery constraints and thermal issues, which affect the performance and efficiency of artificial intelligence (AI) processing systems.

Innovation Solution

The proposed solution involves an integrated circuit package design where the memory die is positioned below the compute die or on its sides, utilizing high-bandwidth memory (HBM) and incorporating ferroelectric RAM (FeRAM) or other fast access memories, with a substrate acting as an active interposer to enhance communication and thermal management.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If DRAM is stacked on top of compute die, then memory capacity is increased, but I/O bandwidth is limited due to periphery constraints

Engineering Contradiction:
Improvememory capacityVSAvoidI/O bandwidth
Core Design Contradiction:
Quantity of substanceVSSpeed

Solution Approach 1:

The patent transitions from 2D periphery-based I/O to 3D vertical stacking architecture. Memory dies are stacked vertically above the compute die, with I/O connections established through the stacking interface rather than peripheral edges. This dimensional change allows multiple I/O channels to be formed through the vertical stack, dramatically increasing bandwidth while maintaining compact footprint.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The memory system is segmented into multiple separate memory dies stacked vertically, each die providing independent I/O channels to the compute die. This segmentation allows parallel data transmission through multiple stacked layers, increasing overall bandwidth while each individual die remains manageable in size and complexity.

Inventive Principle:
Principle #1Segmentation

2Quantity of substance

If DRAM is stacked on top of compute die, then memory capacity is increased, but thermal issues arise affecting performance

Engineering Contradiction:
Improvememory capacityVSAvoidthermal management
Core Design Contradiction:
Quantity of substanceVSTemperature

Solution Approach 1:

The patent introduces an interposer substrate as an intermediary layer between the compute die and stacked memory dies. This interposer serves as a thermal management interface, providing pathways for heat dissipation and enabling separate thermal management strategies for compute and memory components while maintaining electrical connectivity.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

Thermal management is segmented by separating compute and memory components into distinct stacked layers. This spatial segmentation allows independent thermal management approaches for each component type, with heat sinks and cooling solutions optimized for their specific thermal characteristics and operational requirements.

Inventive Principle:
Principle #1Segmentation

3Speed

If compute die is perforated with TSVs to couple with memory bumps, then I/O bandwidth is increased, but device complexity increases

Engineering Contradiction:
ImproveI/O bandwidthVSAvoiddevice complexity
Core Design Contradiction:
SpeedVSDevice complexity

Solution Approach 1:

The patent merges the TSV formation process with the memory die stacking process. TSVs are formed in the compute die to align with bump patterns on memory dies, and the stacking process simultaneously establishes both mechanical support and electrical connectivity. This merging reduces overall manufacturing steps and complexity compared to separate processes.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The memory die bump pattern serves a dual function: providing mechanical attachment points and defining the TSV location map for the compute die. The bump arrangement self-determines the TSV configuration, eliminating the need for separate alignment processes and reducing manufacturing complexity while enabling high-bandwidth connections.

Inventive Principle:
Principle #25Self-service

Data Source

PatentUS11844223B1Ferroelectric memory chiplet as unified memory in a multi-dimensional packaging
Publication Date: 2023.12.12 KEPLER COMPUTING INC
  • US11844223B1 patent drawing
  • US11844223B1 patent drawing
  • US11844223B1 patent drawing

AI summary

A ferroelectric memory chiplet in a multi-dimensional packaging. The multi-dimensional packaging includes a first die comprising a switch and a first plurality of input-output transceivers. The multi-dimensional packaging includes a second die comprising a processor, wherein the second die includes a second plurality of input-output transceivers coupled to the first plurality of input-output transceivers. The multi-dimensional packaging includes a third die comprising a coherent cache or memory-side buffer, wherein the coherent cache or memory-side buffer comprises ferroelectric memory cells, wherein the coherent cache or memory-side buffer is coupled to the second die via I/Os. The dies are wafer-to-wafer bonded or coupled via micro-bumps, copper-to-copper hybrid bond, or wire bond, Flip-chip ball grid array routing, chip-on-wafer substrate, or embedded multi-die interconnect bridge.