RISC-V Solution

Scale by adding dies, not by growing one chip

A die-to-die chiplet platform in seven configuration styles, from a CPU-to-memory baseline up to a multi-die AI and HBM mesh. Each style combines WIOWIZ RISC-V and NPU dies over standard interconnects (UCIe, AIB, HBM, CXL and UALink), so systems scale by adding dies instead of growing one large chip.

Architecture

PACKAGE SUBSTRATE RISC-V CPU compute die NPU AI engine die HBM Stack high-bandwidth memory Host Interface CXL and I/O die NPU AI engine die UALink Port scale-out mesh UCIe AIB UCIe AIB UCIe UCIe UALink

Architectural view. Each block is a WIOWIZ chiplet connected over standard die-to-die links, and internal micro-architecture is not shown.

RISC-V Architecture

WIOWIZ RISC-V compute dies and NPU dies are connected over standard die-to-die links, letting a system scale by adding chiplets (more compute, more memory, more nodes) instead of building one very large monolithic die.

Specifications

ConfigurationDescription
Config 1CPU to on-chip memory, die-to-die baseline
Config 2CPU to CPU over a real die-to-die PHY
Config 3CPU to high-bandwidth memory (HBM)
Config 4AI engine (NPU) to high-bandwidth memory
Config 5CPU plus NPU with dual HBM
Config 6CXL memory and I/O expansion
Config 7multi-node AI and HBM mesh (UALink scale-out with all-reduce)
InterconnectsUCIe, AIB, HBM, CXL, UALink
DiesWIOWIZ RISC-V and NPU chiplets
IP ownershipWIOWIZ IP and WIOWIZ VIP
Design flowWIOWIZ in-house EDA
StatusR&D program (seven configurations in bring-up)

Key Features

Seven configuration styles

One platform spans a CPU-to-memory baseline up to a multi-die AI and HBM mesh.

Standard die-to-die interconnects

Chiplets talk over open die-to-die standards (UCIe, AIB) rather than a proprietary link.

High-bandwidth memory integration

HBM stacks sit beside compute dies to feed CPU and NPU chiplets at full rate.

CXL memory and I/O expansion

A host interface die adds CXL memory and I/O reach without redesigning the compute dies.

UALink scale-out mesh

UALink ports join nodes into an all-reduce mesh for multi-node AI and HBM.

RISC-V and NPU chiplets

Every compute and AI die is WIOWIZ RISC-V and NPU silicon, architected in house.

Applications

Datacenter accelerators AI scale-out High-bandwidth memory systems Multi-die SoCs