AI Chip Center for Emerging Smart Systems

What AI Chip Center for Emerging Smart Systems does

ACCESS (AI Chip Center for Emerging Smart Systems) is a Hong Kong-based, university-affiliated R&D center operating under the InnoHK Innovation and Technology Platform, established in 2020-2021 by HKUST with HK$443.9 million in initial government funding. It is positioned as Asia's first transnational R&D consortium for AI chip design, with core academic partners at Stanford University, the University of Hong Kong, and the Chinese University of Hong Kong. The center employs 36 supervising scholars and a target headcount of 60-80 full-time researchers supported by 100+ affiliated academics, executing four research programs spanning computing-in-memory (CIM) architecture, hardware-software co-design for edge AI, EDA tools for large-scale AI hardware, and cross-layer optimization for targeted applications.

The technical portfolio centers on the AC-Transformer chip family — a 28nm AI accelerator supporting both Transformer and CNN natively (ISSCC 2025), the AC-Transformer-RHB 3D Logic+ReRAM chip for large-model deployment, and the AC-Transformer-DS multimodal LLM accelerator supporting DeepSeek/Llama/Gemma/Qwen. Supporting IP includes the AC-Copilot application-algorithm-hardware co-design toolchain, AutoDCIM (world's first automated DCIM EDA tool), AC-CVXPY for convex optimization acceleration, and Ai-DTCO specialty IP generation. Application demonstrators span portable healthcare (retinal screening, ultrasound segmentation at 50 FPS, virtual staining, genome analysis), FinTech (sub-microsecond arbitrage hardware), and robotics (OctopOS embodied AI runtime). Reported efficiency metrics include 16.3x higher computational efficiency and 7.2x energy efficiency improvement over commercial 8nm AI chips for AC-Transformer, up to 26 TOPS/W for DCIM modules, and 301.6x energy efficiency improvement for the portable genome analysis prototype.

Commercially, ACCESS monetizes through three channels: government research grants (InnoHK initial funding plus RAISe+ project funding), technology licensing to semiconductor and EDA enterprises, and technology transfer via four spin-off companies — HarbourTek (edge AI), SiliconRight (AI-DTCO), QuantaEdge (FinTech hardware acceleration), and OCTOP (embodied AI runtime). Pricing and unit economics are not publicly disclosed, reflecting both its research-center posture and case-by-case negotiation of licensing terms. Customer segments include academic collaborators, semiconductor/EDA companies, medical device and healthcare technology firms, FinTech firms, and robotics companies; the center reports no disclosed enterprise revenue figures.

Name

AI Chip Center for Emerging Smart Systems

Legal name

AI Chip Center for Emerging Smart Systems

Website

https://inno-access.hk

Company type

Private

Founded year

2021

Operating status

Operating

Headcount range

11–50 employees

Short description

AI Chip Center for Emerging Smart Systems (ACCESS) is a Hong Kong InnoHK-funded transnational R&D consortium, established by HKUST with Stanford, HKU, and CUHK, developing computing-in-memory AI accelerator chips, hardware-software co-design tools, and EDA automation for edge AI, healthcare, FinTech, and robotics applications via licensing and four spin-off companies.

Ownership category

akta.pro rank

Product category

AI Semiconductor Research and Chip Design

NAICS

Computer Systems Design and Related Services (54151), Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services (5182)

SIC

Electronic Computers (3571), Computer Communications Equipment (3576)

akta.pro primary industry

Storage/IO Controllers & Bridge ICs (USB/PCIe/SATA/NVMe) (HDAHAJAI)

Keywords

• AI chip design
• Computing-in-memory architecture
• Hardware-software co-design
• Electronic design automation
• AI accelerator development

Headquarters

HQ city

Hong Kong

HQ country

Hong Kong SAR China

HQ region

Asia

Offices1 record

Markets served

GTM type

B2B

Offering type

Services

Cost components

Personnel, Technology or R&D, Infrastructure, Operations, Marketing or Sales

Revenue model

1. Government Research Funding (InnoHK): ACCESS is funded through Hong Kong SAR Government InnoHK Innovation and Technology Platform initial grant of HK$443.9 million. Additional funding secured through RAISe+ (Industry-Academia-Research 1+ Scheme) for specific research projects.
2. Technology Transfer and Spin-off Companies: ACCESS generates revenue through technology transfer and commercialization of research outcomes via spin-off companies including HarbourTek, SiliconRight, QuantaEdge, and OCTOP, which develop and commercialize ACCESS's AI chip technologies.

Go-to-market motion1 record

Distribution channels4 records

Marketing channels9 records

Core offering

ACCESS is a transnational R&D consortium that designs and develops proprietary AI chip technologies including AI accelerator chips (AC-Transformer family), computing-in-memory architectures, 3D-stacked ReRAM chips, and electronic design automation tools (AC-Copilot, AutoDCIM). It commercializes research through technology licensing, industry partnerships, and four spin-off companies targeting edge AI, semiconductor manufacturing, embodied robotics, and high-frequency trading applications.

Product overview

ACCESS (AI Chip Center for Emerging Smart Systems) is a multi-disciplinary research center under Hong Kong's InnoHK initiative, developing AI chip technologies through four research programs. The core product portfolio centers on AC-Copilot (co-design toolchain) and AC-Transformer family of AI accelerator chips, complemented by specialized solutions including DCIM technology for in-memory computing, ultra-high energy efficiency accelerators for healthcare devices, genome analysis systems, ultrasound image segmentation, virtual staining for medical imaging, adaptive depth processors, and convex optimization solvers. The portfolio includes AC-Transformer (28nm CNN/Transformer chip), AC-Transformer-RHB (55nm 3D Logic+ReRAM chip), and AC-Transformer-DS (multimodal large model accelerator), all designed through the AC-Copilot co-design workflow and toolchain.

Differentiator

Problem solved

Functional benefit

Brands

• AC-Copilot: Application-Algorithm-Hardware Co-design Workflow and Toolchain for AI accelerator design
• AC-Transformer
• AC-Transformer-RHB
• AC-Transformer-DS
• Ultra-High Energy Efficiency AI Accelerator
• DCIM Genome Analysis Prototype
• AC-CVXPY
• Ai-DTCO

Products and services

• AC-Copilot Integrated design toolchain for AI chip development that integrates neural architecture search, model compression, hardware simulation, and software generation. Supports mainstream deep learning frameworks (PyTorch, ONNX), hardware architecture design, and SDK deployment. Targeted at AI chip designers and semiconductor companies seeking end-to-end co-design automation.
• AC-Transformer High power-efficiency AI accelerator chip supporting native Transformer and CNN dual architectures, fabricated at TSMC 28nm with peak performance of 2.56 TOPS/s. Features sparse attention computation, non-linear operator hardware fusion (18 operation types), and intelligent memory scheduling. Targeted at AI inference applications in autonomous driving, robotics, and edge AI systems.
• AC-Transformer-RHB 3D-stacked Logic + ReRAM chip using CMOS and ReRAM integration technology for large language model deployment. Achieves 75-80% external memory access compression via near-memory processing, low-bit quantization, and speculative decoding. Manufactured at 55nm node with 8MByte ReRAM capacity and 25.6GByte/s ReRAM bandwidth.
• AC-Transformer-DS AI accelerator chip for multimodal large models supporting 40+ core instructions with scalable framework. Supports Large Language Models (DeepSeek, Llama, Gemma, Qwen), Visual Language Models (Llava, Qwen-VL, DeepSeek-VL), and generative models (Grounding DINO, Stable Diffusion). Targeted at multimodal AI applications requiring efficient inference.
• Ultra-High Energy Efficiency AI Accelerator for Portable Healthcare Devices Portable AI retinal disease screening device with ultra-high energy efficiency AI accelerator for diabetes and hypertension patients. Algorithm-architecture co-design achieves 7x energy efficiency improvement over edge GPUs and reduces off-chip memory access by over 80%. Targeted at portable medical device manufacturers and healthcare providers.
• Digital Compute-In-Memory (DCIM) Design and Applications Digital Compute-in-Memory technology integrating storage and computing logic at the physical layer, including AutoDCIM—the world's first EDA tool for application-specific DCIM chip generation. AC-DCIM module achieves up to 26 TOPS/W computational efficiency using 28nm process. Targeted at semiconductor companies developing energy-efficient AI accelerators.
• DCIM-enabled Portable Genome Analysis Prototype Portable genome analysis prototype using DCIM technology enabling efficient, low-cost genetic testing in resource-limited environments. Features PoliFormer software architecture and achieves 5.61x throughput acceleration and 301.6x energy efficiency improvement over existing chips. Targeted at genomics researchers and portable diagnostic device developers.
• Real-Time Ultrasound Image Segmentation AI Accelerator Real-time cardiac image segmentation system processing ultrasound imaging at 50 FPS with Dice similarity coefficients of 91.8% (LV), 89.6% (LA), 84.3% (Myo) for cardiac function evaluation and pathology detection. Built on AC-Codesign-v1 chip architecture. Targeted at portable ultrasound device manufacturers and cardiology AI developers.
• Virtual Staining for Fast Histological Imaging Fast histological imaging system combining UV-PAM with deep learning algorithms to generate virtual H&E stained images from untreated fresh tissue samples. Built on Xilinx U250 FPGA with 1.6 TOPS peak performance using 8 AI processing cores. Developed in collaboration with PhoMedics for the CHAMP microscope. Targeted at surgical oncology and digital pathology applications.
• Adaptive Monocular Depth Processor Depth perception system fusing Vision Transformer with Time-of-Flight (ToF) sensor data using self-adaptive fusion engine with local compensation and incremental gradient fusion algorithm. Requires only single infrared input, eliminating multi-sensor calibration. Targeted at robotics, autonomous driving, and 3D sensing applications.
• AC-CVXPY Automated algorithm-hardware co-design framework for convex optimization solvers. Achieves 30.5x geometric mean speedup and 127.0x energy efficiency improvement over equivalent CPU backend on 100 QP problems across 5 application domains. Targeted at robotics control systems, high-frequency trading firms, and optimization-intensive applications.
• AI-DTCO and Specialty IP AI-based Design-Technology Co-Optimization (DTCO) solution that co-analyzes circuit and process technology characteristics. Automatically generates optimal specialty IP for high-performance computing, high reliability, and ultra-low-power applications. Uses embedded sensors to monitor IC manufacturing process variations and provide design optimization guidance. Targeted at semiconductor companies and IC design teams.

Quantifiable outcome

• AC-Transformer achieves 16.3x higher computational efficiency and 7.2x energy efficiency improvement compared to commercial 8nm AI chips
• +7 more outcomes

Named customers3 records

Segments5 records

Ideal customer profiles5 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

AI capability19 records

Feature11 records

Partnerships

14 partnerships are on record, tiered strategic, core and minor.

• Korea Advanced Institute of Science and Technology (KAIST)strategicStrategic or Co-development Partner · 8 December 2025KAIST delegation visited ACCESS for academic exchange on data analysis algorithms and FPGA neural network accelerators. The visit included faculty, postdoctoral researchers, PhD and Master's students. Discussions covered ACCESS's genome data analysis algorithm acceleration architecture and FPGA neural network accelerator generation technologies.
• Science Tokyo (Tokyo Institute of Technology)strategicStrategic or Co-development Partner · 6 December 2024Joint delegation visit with KAIST to ACCESS for academic exchange. Prof. Daichi Fujiki from Science Tokyo's AI Computing Research Unit participated, sharing research perspectives with ACCESS researchers on AI algorithm and accelerator co-design topics.
• École Polytechnique Fédérale de Lausanne (EPFL)strategicStrategic or Co-development Partner · 22 November 2024Prof. Giovanni De Micheli, Director of Integrated Systems Laboratory and Scientific Director of EcoCloud Center at EPFL, visited ACCESS and engaged in discussions on potential future collaborations. Prof. De Micheli is renowned for his contributions to automated design of Network on Chip. The visit fostered valuable discussion on EPFL-ACCESS collaboration opportunities.
• Hong Kong Applied Science and Technology Research Institute (ASTRI)coreStrategic or Co-development Partner · 8 June 2022ACCESS and ASTRI signed an MoU to jointly promote the development and commercialization of AI chips in Hong Kong. The partnership combines ACCESS's academic research capabilities with ASTRI's applied research infrastructure and industry channels to accelerate technology transfer of AI chip research outcomes. The first collaboration project focuses on promoting specialized AI chips in the Smart IoT (AI-IoT) domain.
• Hong Kong University of Science and Technology (HKUST)coreStrategic or Co-development PartnerHKUST is the host university and primary founder of ACCESS, providing institutional backing and research infrastructure. HKUST has invested significantly in establishing ACCESS as a wholly university-owned R&D center under InnoHK. ACCESS operates in close coordination with HKUST's School of Engineering and Department of Electronic and Computer Engineering.
• Stanford UniversitycoreStrategic or Co-development PartnerStanford University is one of the four founding collaborating universities of ACCESS, joining HKUST, University of Hong Kong, and Chinese University of Hong Kong to form Asia's first transnational AI chip design R&D consortium.
• University of Hong Kong (HKU)coreStrategic or Co-development PartnerUniversity of Hong Kong is one of the four founding collaborating universities of ACCESS, contributing faculty and researchers to the transnational AI chip design R&D consortium.
• Chinese University of Hong Kong (CUHK)coreStrategic or Co-development PartnerChinese University of Hong Kong is one of the four founding collaborating universities of ACCESS, contributing faculty and researchers to the transnational AI chip design R&D consortium.
• HKUST (Guangzhou)coreStrategic or Co-development PartnerHKUST (Guangzhou) is part of the dual-university network leveraged by ACCESS. Through the Blue Bay Entrepreneurship Program platform, ACCESS connects technological R&D talents, entrepreneurial teams, venture capital firms, and industrial partners across Shenzhen and Hong Kong.
• Hong Kong Science and Technology Parks Corporation (HKSTP)coreStrategic or Co-development PartnerHKSTP co-organizes ACCESS Technology Symposium and provides the Hong Kong Science Park facilities where ACCESS headquarters are located. HKSTP facilitates the center's presence in Hong Kong's innovation ecosystem.
• HKUST Blue Bay Incubator HarborstrategicChannel Partner/ Reseller/ DistributorACCESS is establishing presence at HKUST Blue Bay Incubator Harbor in Shenzhen Qianhai. This is the first incubator in Shenzhen independently operated by a Hong Kong higher education institution. The platform pools technology, entrepreneurial talents, venture capital, and industrial resources from Shenzhen and Hong Kong to build a cross-border innovation ecosystem.
• Hong Kong Trade Development Council (HKTDC)minorGTM or Marketing PartnerFirst collaboration with HKTDC to co-host the 'Under 30 Series - Meeting Young Research Scientist' seminar at the Hong Kong Electronics Fair and International Electronic Components and Production Technologies Exhibition.
• Hong Kong C. K. James Robotics Institute (CKSRI)strategicStrategic or Co-development PartnerCollaboration with HKUST's CKSRI on Smart Rehabilitation Robotic System. The system is powered by ACCESS's AC-Transformer AI chip and enables real-time analysis of human dynamics and posture for rehabilitation treatment precision improvement.
• PhoMedicsstrategicStrategic or Co-development PartnerPhoMedics collaborated with ACCESS to customize the AI accelerator core of the AC-Codesign-v1 platform for the CHAMP microscope. The customized solution is built on Xilinx U250 FPGA platform for the operating room environment, achieving 1.6 TOPS peak performance for virtual staining of fresh tissue samples during cancer surgery.

Scale indicators8 records

Recent moves8 records

Expansion highlights5 records

Regional players

• Horizon Robotics: China-based AI chip company specializing in autonomous driving processors. Comparable given ACCESS's autonomous driving applications and Chinese ecosystem footprint via InnoHK.
• Cambricon Technologies: China's leading AI chip company for cloud and edge inference, listed on Shanghai STAR Market. Comparable as a regional AI chip designer, though commercially much further along.

Broad incumbents

• Synopsys: Largest EDA tools and semiconductor IP vendor globally, including AI/ML-enhanced design automation. Comparable given ACCESS's development of AI-driven EDA tools (AC-Copilot, AutoDCIM, Ai-DTCO) targeting similar workflows.
• SiFive: RISC-V chip IP company licensing processor IP to semiconductor companies. Comparable given ACCESS's participation in RISC-V Summit China and IP-licensing commercialization approach.

Direct peers

• SambaNova Systems: AI accelerator chip company with reconfigurable dataflow architecture for large model inference and training. Comparable as a fabless AI chip designer serving enterprise and government customers with custom silicon.
• Groq: AI inference chip company developing LPU (Language Processing Unit) accelerators. Comparable as a fabless AI chip designer targeting low-latency inference, though Groq uses deterministic tensor streaming architecture rather than CIM.
• Tenstorrent: AI chip company building RISC-V-based AI accelerators and licensing IP to other chip companies. Comparable given ACCESS's RISC-V ecosystem participation and IP-licensing commercialization model.
• Cerebras Systems: AI compute company using wafer-scale integration for AI training and inference. Comparable as an AI chip innovator focused on memory-bandwidth bottleneck solutions, though using different architectural approach.

Emerging players

• Syntiant: Edge AI chip company focused on ultra-low-power neural decision processors for always-on devices. Comparable given ACCESS's ultra-low-power edge AI positioning in healthcare and IoT verticals.
• Mythic AI: Analog compute-in-memory AI processor company for edge AI inference. Closely comparable architecturally to ACCESS's CIM focus, though Mythic has struggled commercially.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat4 records

Key risks6 records

Key highlights6 records

Customer concentration

• LinkedInlinkedin.com/company/inno-accesshk

Revenue estimate

Valuation estimate

Number of profiles

Profiles4 records

Subsidiaries4 records

Funding overview

Funding rounds

Investors

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about AI Chip Center for Emerging Smart Systems