• On-device inference
• Real-time response
• Operates offline or with intermittent networks
Latency & Cloud Dependency
Centralized cloud inference introduces latency, limits responsiveness, and requires continuous connectivity.
Ultra Low Latency AI
• RISC-V + 16/12nm = optimal cost/performance
• Lyra for high-volume terminals
• Hyper for premium nodes
Latency & Cloud Dependency
Centralized cloud inference introduces latency, limits responsiveness, and requires continuous connectivity.
Efficient Edge Compute
• Local data processing
• Better compliance with GDPR, HIPAA, etc.
• Supports secure boot, encryption
Latency & Cloud Dependency
Centralized cloud inference introduces latency, limits responsiveness, and requires continuous connectivity.
On-Device Security & Privacy
• Sub-10W power envelope
• Passive cooling
• Compact FCBGA package for panel integration
Latency & Cloud Dependency
Centralized cloud inference introduces latency, limits responsiveness, and requires continuous connectivity.
Compact, Low-Power Design
Our Solutions
Technologies
Real-Time Edge AI Processing
• Dedicated RISC-V cores with INT8/FP16 AI
acceleration
• On-device vision, voice, and language inference
• TVM / ONNX Runtime optimized pipelines
Scalable Compute Architecture
• Hyper: High-performance multi-core for immersive
AI terminals
• Lyra: Cost-optimized for dense deployments
• Shared SDK for consistent development experience
Compact & Passive-Thermal Design
• Sub-10W power envelope
• Passive cooling, no fan/heatsink required
• FCBGA packaging for slim panel integration
AI-First Display Engine
• Supports up to 2K@60Hz or FHD+@90Hz
• Local rendering with adaptive refresh & compression
• Integrated display and video codec pipeline
Developer-Ready Platform
• Secure Boot, eFuse, hardware crypto engine
• Fully local AI model execution (no cloud leakage)
• Meets data residency and compliance demands
Developer-Ready Platform
• Unified BSP for Linux Yocto / AOSP
• Model portability across Nova & Hyper
• SDK-ready for NLP/CV/ASR applications at the edge
Hyper Series
Computing
System
CPU
RISC-V, 4-8x Performance cores (2.2–2.4GHz) + 2-4x Efficiency cores (1.5–1.8GHz)
GPU
200+ GFLOPS FP32, OpenGL ES 3.2 and Vulkan 1.2 capable
AI
5-8 TOPS (INT8), 2+ TOPS (FP16)
DSP
Dual-core, capable of 1 TOPS CV pre-processing or 200 GOPS audio DSP
Memory & Storage
LPDDR4X / LPDDR5, 32-bit interface, up to 8GB
UFS 2.1 or eMMC 5.1
Multi-Media
Video Codec : 4K@30fps decode (H.265/H.264), 1080p@30fps encode
I/O Interfaces
PCIe Gen3 x1, USB 3.1/3.2, MIPI-CSI/DSI (4+4 lanes), SDIO, I²S, I²C, SPI
Display
Up to 2560×1600 @ 60Hz, or FHD+ @ 90Hz (Dual Panel capable)
Security
Secure Boot, AES-256, SHA-2, hardware RNG, eFuse
Applications
Smart dash, premium AI panels, AI terminals
Product Family
Hyper 11 : 2025Q4 tapeout, 2026Q1 ES, 2026Q3 MP Hyper 2x~4x : 2026Q2 tapeout
Lyra Series
Computing
System
CPU
4–6-core RISC-V: In-Order Cores (up to 2.0GHz)
GPU
50–80 GFLOPS FP32, supports OpenGL ES 2.0 / 3.0
AI
1.5–3 TOPS (INT8), FP16 optional
DSP
50–100 GOPS, audio-focused, basic vision filter support
Memory & Storage
LPDDR4 up to 4GB, 16/32-bit
eMMC 5.1
Multi-Media
Video Codec : 1080p@30fps H.264/H.265 decode/encode(opt)
I/O Interfaces
USB 2.0/3.1, MIPI-CSI/DSI, SPI, UART, GPIO, I²C, I²S
Display
FHD @ 60Hz (Dual Panel capable)
Security
Basic Secure Boot, AES128, software crypto acceleration
Applications
Entry-level AI display panels, retail kiosks, smart speakers/screens
Product Family
Lyra 111 : 2025Q1 tapeout, 2025Q2 ES, 2025Q4 MP
Lyra 131 : 2025Q1 tapeout, 2025Q4 ES, 2026Q2 MP
Lyra 211 : 2025Q1 tapeout, 2025Q2 ES, 2025Q4 MP
Lyra 212 : 2025Q1 tapeout, 2025Q3 ES, 2026Q1 MP
Lyra 231 : 2025Q1 tapeout, 2025Q4 ES, 2026Q2 MP
Lyra 3xx ~ 6xx : 2026Q1~ tapeout