HBM3 PHY V2 in TSMC (N5, N3E)

Overview

The Synopsys HBM3 PHY is a complete physical layer IP interface (PHY) solution for high-performance computing (HPC), AI, graphics, and networking ASIC, ASSP, and system-on-chip (SoC) applications requiring high-bandwidth HBM3 DRAM interfaces operating at up to 9600 Mbps. The synopsys HBM3 PHY offers superior power efficiency compared to any other off-chip memory interface and supports up to 4 active operating states enabling dynamic frequency scaling. To minimize area, the PHY utilizes an optimized micro bump array. Support for longer channel lengths allows more flexibility in the PHY placement on the SoC without impacting performance. The PHY provides a complete HBM3 interface solution when combined with Synopsys’ HBM3 Controller IP and HBM3 memory model VIP.
The configurable Synopsys HBM3 PHY is provided as a set of hard macrocells delivered as GDSII. These hard macrocells include integrated application- specific HBM3 I/Os required for HBM3 signaling. The design is optimized for high performance, low latency, low area, low power, and ease of integration.
The hard macrocells are easily assembled into a complete 1024-bit HBM3 PHY. The RTL-based PHY Utility Block (PUB) supports the GDSII-based PHY components and includes the PHY training circuitry, configurations registers
and BIST control. The HBM3 PHY includes a DFI 5.0-compatible interface to the memory controller, supporting DFI 1:1:2 and DFI 1:2:4 clock ratios. The design is compatible with both metal-insulator-metal (MIM) and non-MIM power decoupling strategies. Synopsys also offers a pre-hardened “drop-in” version of the Synopsys HBM3 PHY for customers that do not have significant custom requirements. For customers that require a custom hard Synopsys HBM3 PHY, Synopsys also offer PHY hardening design services.

Key Features

  • Low latency, small area, low power
  • Compatible with JEDEC standard HBM3 DRAMs
  • Data rates up to 9600 Mbps
  • 8H, 12H and 16H HBM3 DRAM stacks supported
  • Support for fast switching between up to 4 frequencies
  • Product subcomponents designed to precisely control timing critical delay and skews
  • Controller DFI-compatible interface (DFI v5.0 Addendum 2)
  • Includes 1 phase-locked loop (PLL) per PHY and digital delay lines necessary to meet timing specifications
  • Separate transmit DQ and transmit DQS as well as transmit Row/Col and transmit CK delay lines
  • Includes separate receive DQS delay lines for both rising and falling edge of DQS
  • Supports 32 pseudo-channel HBM3 DRAM systems
  • Boot time impedance calibration
  • Programmable I/O drive strength matching the HBM3 DRAM
  • Delay line VT compensation
  • Time axis data eye training
  • PHY VREF can either be supplied externally through a bump or use an internal VREF generator that can provide a programmable VREF for use internal to the PHY
  • Physical implementation of the top-level HBM PHY hard macro is designed to be compatible with a face centered rectangular (FCR) micro bump pattern similar to the HBM3 micro bump pattern
  • Area optimized micro bump pattern for smaller PHY area
  • Enhanced power savings support that includes:
  • Arm® AMBA ® APB interface for configuration register access
  • Test data register (TDR) interface for configuration register access
  • Multiple test modes
  • At-speed loopback testing on both the address and data channels
  • Pre-hardened HBM PHYs are available in select process technologies
  • Services available for custom PHY hardening requirements
  • 2.5D Interposer reference designs available

Benefits

  • Supports 2.5D-based JEDEC standard HBM3 DRAMs with data rates up to 9600 Mbps
  • 16 independent 64-bit memory channels
  • Pseudo-channel operation supported to enable up to 32 32-bit pseudo-channels with 1024 bit PHY
  • Supports up to 4 trained frequencies with <5us switching time
  • DFI 5.0-compatible controller interface
  • PHY independent training capability
  • Comprehensive set of design-for-test (DFT) features

Applications

  • Data center and networking
  • High performance computing (HPC)
  • Artificial intelligence (AI)
  • High-end graphics

Deliverables

  • Executable .run installation file, including GDSII, LEF Files, LVS netlists, .lib/.db timing models, Verilog model, DRC/LVS log files, I/O IBIS model, I/O HSPICE netlist, parameterized Verilog top-level PHY netlist files, sample verification environment, PHY data book, physical implementation guide, application notes, verification guide, installation guide, implementation checklist
  • The PHY Utility Block includes Verilog code, synthesis/STA constraints and scripts, sample verification environment, data book

Technical Specifications

Foundry, Node
TSMC N5, N3E - EFF, FF
Availability
Contact the Vendor
TSMC
Pre-Silicon: 3nm , 5nm
×
Semiconductor IP