💾

Memory Technology and Its Impact on the Industry

Jul 14, 2024,

Memory Technology and its Impact on the Semiconductor Industry

Introduction

  • Memory technology is critical for CPUs, GPUs, and SoCs.
  • Current memory technology is facing significant issues.
  • New memory technology might solve these issues.

Historical Context

  • Transistors have decreased in size since their invention in 1947.
  • Early transistors were centimeters in size; now, they are produced at nanometer scale.
  • Examples: Qualcomm Snapdragon X Elite chip (4nm), M4 chip (3nm).

Memory Scaling Issues

  • Memory, particularly SRAM, has stopped scaling effectively.
  • Historically, SRAM was the preferred fast memory (e.g., for cache).
  • SRAM cells consist of latches built with 4-6 transistors.

Technical Details

  • SRAM vs DRAM: SRAM uses latches and is more power-efficient.
  • SRAM Performance: High performance, stores data close to processing cores.
  • Clock Speed: Data accessed in the range of 250-500 picoseconds.

Industry Trends

  • Increasing memory in chips by Intel, AMD, Nvidia, Apple.
  • SRAM’s inability to scale is problematic due to its integration on the chip die.
  • SRAM doesn’t scale as effectively as logic, leading to increased chip area and cost.

Key Challenges

  • SRAM cell structure doesn’t conform to normal logic design rules, making them sensitive to manufacturing variations.
  • Transition from FinFET to Gate-all-around transistors presents new challenges.

Current Solutions and Innovations

  • Chiplets: Stacking memory on the CPU die for increased capacity and performance.
    • Example: AMD’s V-cache technology uses TSMC’s 3D SoIC packaging tech.
  • 3D Stacking: Mixing different process nodes for core logic and memory to save costs and improve performance.

Emerging Memory Technologies

  • Alternative memory types: MRAM, FeRAM, RRAM, PCM.
  • Phase-Change Memory (PCM): Uses chalcogenide in a superlattice structure.
    • PCM toggles between crystalline and amorphous states using high current pulses.
    • New PCM technology (GST467) developed by Stanford is promising.

Benefits of PCM

  • Small access times (a few nanoseconds), low operating voltage.
  • Smaller dimensions (0.016 micrometer square), more area-efficient than SRAM.
  • Nonvolatile memory with potential for multi-bit storage (limited to specific applications).

Challenges for PCM

  • Integration with current CMOS processes.
  • Reducing programming currents and improving reliability.

Future Directions

  • Ongoing research and innovations to address SRAM limitations.
  • Potential new memory technologies initially targeting DRAM and then moving to L3 cache.
  • AI applications increasing the demand for effective memory solutions.
  • SRAM likely to remain essential for L1 and L2 caches in the coming decades.

Closing Remarks

  • Stay updated on microchip technology trends.
  • Check out and support the channel via the new ASUS Vivobook S 15.

Product Plug: ASUS Vivobook S 15

  • First Copilot+PC laptop based on Qualcomm Snapdragon X Elite chip (4nm).
  • Features: NPU for 45 TOPS, dedicated AI assistant, various AI-powered applications.
  • Highlights: 16 GB RAM, 3K OLED display, 1 TB SSD, multiple ports, and 18 hours battery life.

This summary covers the key points from the presentation, highlighting the issues with current memory technology, emerging alternatives, and the specifics of phase-change memory (PCM).

Full transcript