CLTECH

Leading the Future of Al With TPU-Driven Al Accelerator

Unmatched Computing speed Born for AI/ML Less Energy Consumption 1024 Interconnected Chips

About Us

CL Tech was founded in 2018 by Yifan Yanggong, former core Google TPU R&D engineer, alongside Al hardware and software experts from global tech leaders including Google, Microsoft, and Samsung. Centered around our self-developed TPU architecture high-performance Al chip CHANA? and computing cluster Taize?, our team is building a computing foundation for large model operations that is fully proprietary, independently controllable, and matches leading global standards. Today, CL Tech has been collaborating with governments, telecom operators,and enterprises nationwide to establish multiple ultra-large intelligent computing centers.

CATALOG

Introduction

Core Team
Development
Intellectual Property 0
Business Growth
Investors
Accreditations&Honors

Al Computing Hardware&Solutions

TPU Architecture Al Chip
Al Servers& Intelligent Computing Clusters
Case Studies

Large Model Products & Solutions

Pre-trained Foundation Model Finance Large Model Education Large Model Healthcare Large Model

Company Introduction

Founded in 2018, CL Tech is dedicated to developing high-performance TPU Al accelerators and intelligent computing clusters that support ultra-large-scale Al model computation to empower clients with their Al innovation

Core Team

Founder and CEO: Yifan Yanggong

IndustryLeadership

With 10 years of experience in the field of high-end chips,he has successfully taped out more than ten times in his industrial career, and has mastered expertise spanning advanced processes from 28nm to 7nm,leading the team to achieve success from architectural design and tape-out to client delivery

Google

Ex-Goolge TPU core R&D enginee for the design and development of TPU 2/3/4

ORACLE

Led the design and development of 12 high-performance CPUs, including SPARC T8/M8,at Oracle

A Team of Global Al chip & Algorithm Experts

CL Tech's core team consists of Al hardware and software design experts from leading tech companies such as Google, Microsoft, Oracle, Samsung, NVIDlA, Amazon,and Facebook.The full-stack technology echelon covers various technical fields such as chip design,circuit design,software stack research,system architecture,large model algorithms. Our R&D personnel account for more than 7 0 %

Development

With core founding team formed, CL Tech definied the high-compute GPTPU chip.The company established full chip design and prototype verification teams, completing modular chip design

Headquartered in Hangzhou with secured angel funding, CL Tech built comprehensive chip-level software teams and non-technical support units

2018
2020
2021
2022
2023
2024

Following Pre-A funding, the company finalized CHANA® chip design verification and achieved ZHURONG? inference SoC implementation

Post-Series A funding, CL Tech completed CHANA? SoC design while successfully taping out and industrializing ZHURONG@

With Pre-B/Pre-B+ funding, CHANA? entered mass production and delivery. CLTech achieved ?480M revenue and ?80M profit while recruiting top talent forlarge-model algorithms

After Series B funding, CHANA?-powered computing hardware saw widespread commercialization.CL Tech reached nearly \yen 600 Mrevenue with ?86M profit

Industry-Leading Intellectual Property

All CL Tech products are 1 0 0 % fully self-developed.The company possesses industry leading core intellectual property systems and accreditations

\*Note:The above patents and qualifications are as of June 2025 and continue to grow rapidly

Business Growth

Investors

Accreditations& Honors

、 20+National/provincial/municipal Certifications

1 30+ Technology Innovation& Entrepreneurship Awards

AICOMPUTINGHARDWAREPRODUCTSANDSOLUTIONS

Al Computing Hardware Products And Solutions

Centered around our self-developed TPU architecture high-performance Al chip CHANA? and computing cluster Taize@. CL Tech is building a computing foundation fQr large model operations that is fully proprietary, independently· controllable, and matches leading global standards

CHANA?:TPU AI Accelerator

Fully Proprietary & Independently Controlled

Purpose-Built for Large Model

Developedover ^ { 4 + } years,CHANA? features a fully self-developed IP core, custom instruction set and dedicated computing platform

Equipped with reconfigurable multilevel memory,unified memory-compute architecture,and pipelined spatiotemporal mapping

Industry-Leading Energy Efficiency

CHANA? consumes only 7 0 % the energy of traditional GPUs,with per-compute-unit costsatmerely 5 0 % of Global Leading GPU

Unparalleled ComputingPerformance

204 TFLOPS in TF32 precision,1.5x faster than global leading GPU,while reducing energy consumption by 3 0 %

#

1,024-chip High-Speed Interconnect

Supports up to 1,024 interconnected chips within one cluster, dozens of times greater cluster performance than GPUs

3-4x Performance Leap at ComparableProcessNodes

Built on a TPU architecture born for Al/ML,along with Chiplet technology and 2.5D packaging,3-4x performance leadp at thesame process node

Chip Performance:CHANA?VS Global Leading GPU

GPTPU:Born for AI/ML

TPU architecture is custom-designed for scientific computation, especially Al/ML with a focus on tensor operations for faster,power-efficient performance.This makes TPUs ideal for large-scale deployment in data centers dedicated to deep learning

Optimized Multi-Dimensional Compute Units

Unlike scalar CPU and vector GPU units, TPUs employ multi-dimensional units unrolling convolution operations to maximize data reuse and reduce transfer costs,enhancing acceleration efficiency

Efficient Data Transferand Compact Control

TPUs address deep learning's "memory wall" with aggressive data transfer strategies and compact control units, freeing up more on-chip space for memoryand computation

Al-Focused, Scalable Design

With streamlined control,a custom instruction set,and scalable architecture, TPUs are highly efficient for deep learning and ideal for ultra-large-scale Al training

Taize? Al Server

The Taize@ Al serveris built with 8 CHANA® chips per compute unit.It delivers superior performance,low power consumption,and ease of deployment and maintenance, positioning it as a leader in Al training hardwares

Top-Tier Computing Performance

With industry-leading dual-direction chip-to-chip bandwidth and linear cluster scalability, Taize@ supportsstable,high-efficiency communication in large-scale parallel computing clusters

△

Optimized Load Balancing and High-Performance Computing

Equipped with a 10oG Ethernet interface, Taize@ deeply optimizes data and model parallelism with refined resource scheduling to dynamically balance workload and storage resources

Industry-Leading Energy Efficiency

Taize? uses only half the energy of traditional GPUs for the same Al training workloads,setting a benchmark for high-efficiency,low-power, green computing in the intelligent computing field

Easy, Reliable,and Highly Compatible

Built-in plug-and-play platform management software,secure data encryption,identity verification, continuous monitoring,and seamless compatibilitywith mainstream largemodels

Taize? Large-Scale Al Computing Cluster

Unmatched computing Power

Taize@ Al computing clusters,powered by CHANA? chips, deliver over 400P FLOPS(TF32) per cluster. Supports massive Al models with hundreds of billions of parameters

品

Elastic Scalability

1,024-chip high-speed interconnect enables linear scaling. Outperforms traditional GPU clusters by dozens of times in system-level performance

Intelligent Resource Orchestration

Real-time multi-user workload balancing with dynamic allocation of computing power and storage.Ensures efficient collaboration across complex scenarios

Multi-Scenario Model support

It supports various high-intensity Al computing scenarios,including AlGC large model computing,advanced unmanned driving model training, precise protein structure prediction, AlphaGo,AlphaZero, etc

T

Rapid Deployment

Automated deployment tools slash cluster setuptime.Dedicatedteamsenableagile execution with flexibleworkflows

Enterprise-Grade Security

Advanced safeguards against unauthorized access and malicious attacks.Guarantees end-to-end data protection

Client Case:Shenzhen Unicom

In September 2024, CL Tech and Shenzhen Unicom launched Guangdong's first Al computing center fully powered by domestic TPU chips.This training-inference integrated hub anchors Unicom's core intelligent computing infrastructure in Shenzhen.

CL Tech deployed its Taize@ Al computing clusters with self-developed CHANA@ TPUs, establishing a robust Al infrastructure foundation.

Integrated with Unicom Cloud's "Xingluo" platform,the center adapts heterogeneous computing resources through unified scheduling of GPGPU, Al,and HPC workloads.It delivers full-stack computing services for financial, educational,and healthcare Al applications,enhancing Shenzhen's industrial intelligence transformation.

Kick-off Ceremony

Client Case: Zhejiang University

To address Zhejiang University's growing high-performance computing needs, ZJU partnered with Taiji Co., Ltd.and CL Tech to deploy Taize@ Al servers powered by CHANA? TPUs.These high-performance, energy-efficient,and compatible systems now accelerate research and teaching across multiple ZJU projects.

Research&Application of Machine Learning Course Image Processing Technology Application
Large-scale Simulation Scenarios Data Science Course
Experiments Training&Optimization of Deep Learning Models

Economic Value

Social Value

This project eliminates universities' dependence on imported high-performance computing systems, ensuring sovereign control of computing resources. It reduces Al computing costs by over 5 0 % versus imported servers while easing financial burdens.Researchersaccelerate large-scale data processing, complex scenario simulation, and theoretical validation, shortening research cycles and accelerating innovation

As innovation hubs, universities' adoption of domestic chipsboosts industry confidence and provides vital real-world feedback for technological advancement. Deep industry-academia collaboration develops talent for sovereign, high-quality IT growth, catalyzing domestic chip adoption

Client Case: Shangrao Xingluo 1000-chip Intelligent Computing Center

Jointly invested by Xingluozhisuan (hangzhou） Technology Co., Ltd一a wholly-owned subsidiary of listed firm Airbluer [30l259]—and Shangrao High-Speed Rail Economic Zone Administrative Committee.As Jiangxi Province's first thousand-chip-scale Al center fully powered by CL Tech's GPTPU chips,it addresses regional high-performance computing demands and drives local Al industrial growth.

Phase 1 deploys 128 Taize@ GPTPU Al servers housing 1,024 CL Tech's self-developed CHANA GPTPU chips.The cluster delivers over 400 PetaFLOPS (TF32) computing power with sparsity acceleration, providing sovereign computing resources for regional Al applications.

Project Value

Delivered on schedule in late 2024, this computing center efficiently supports large-scale Al workloads including AIGC model training and inference. Serving domestic Al enterprises, it accelerates cross-industry Al deployment while propelling Shangrao's digital economy and Al industrial advancement.

Large Model Products And Solutions

CL Tech builds proprietary pre-trained models with industry retraining/inference interfaces,powering finance,healthcare, media, marketing, education, and programming. Through our domain-specific large model development platform, we foster an application ecosystem transforming industrialapplications

Pre-trained Large Models

CL Tech is committed to creating foundational Al models that deliver "rigid" output, addressing common issues in generic models like poor targeting, high hallucination rates,and low content relevance. By integrating domain-specific knowledge bases, CL Tech aims to build models ready for direct deployment. These self-developed, pre-trained models include retraining and inference interfaces tailored to sectors like finance, healthcare,media,marketing,education,and programming

Current Generic Al Models Industry Needs for Al Models
四 团 Q具 X弧
Soft Dialogue VS Rigid Problem-Solving Strong multi-turn conversational ability Natural language understanding forprecise issue resolution Lacks specialized knowledge Structured and reliable responses for professional questions to reduce uncertainty Limited accuracy in Professional knowledge for exact industry-specific problem-solving quantitative solutions

GI LUI

From GUI to LUI

Shifting from graphic (GUl) to language-based (LUl) interfaces for lightweight, efficient data interaction

DeepDatabase Penetration,Avoiding Generic Responses Integrating structured data to penetrate databases and core infrastructure, enabling precise business logic mapping for streamlined, business-focused insights

Balanced Rigid Models

Built on robust understanding and Q&A capabilities,these models combine flexible interactions with reliable,high-impact problem-solving,designed for easy deploymentand high trustworthiness

Finance Large Model

CL Tech co-created "Taize@ Lianqi" with China's top financial institutions.This industry first rigid model for financial data processing delivers structured data analysis, automated visualization,advanced processing,and predictive modeling capabilities

Application Scenarios

CL Tech enables precise deployment of its finance large model for specialized scenarios like stock/fund selection, powering next-generation Al-driven advisory applications. The solution has been validated with China's leading financial institutions for NL2SQL implementation and market intelligence analysis

Education Large Model

CL Tech collaborates with leading listed education enterprises to co-develop the "Taize@ Zhuji" Education Large Model,advancing teaching efficiency, quality,and personalized intelligent education from learning environments,resource optimization,and educator roles

Innovative Low-rank Adaptive Aynamic Matrix Technology Instantaneous Proximal Discrete Attention Optimization The solution has been validated with leading listed education
Application enterprises for constructing disciplinary knowledge graphs,
Scenarios optimizing learning pathways, and enhancing assignment/ examassessment systems,among other educational applications

Healthcare Large Model

CL Tech's self-developed "Taize@ Jindan" healthcare large model constructs Chinese medical fine-tuning instruction sets through authoritative knowledge processing, enables private deployment,and safeguards medical/biological data security

Innovative Technology Application Scenarios

CL Tech has validated its healthcare large model with Zhejiang's top healthcare informatics group across clinical applications including medical QA, EHR analysis and diagnostic report interpretation. Expert evaluations confirm the model outperforms ChatGLMand ChatGPT in drug-indication matching and treatment efficacy assessmentwhile matching ChatGPT's performance in medical knowledge QA and diagnostic reasoning, surpassingother open-sourcealternatives

Follow us Keep updated with the latest news

C
Tel.:
0571-28001676

Email: contact@chipltech.com

。