HyperAccel is a Republic of Korea (South Korea)-based fabless company developing AI chips to enable widespread deployment of Generative AI (GenAI). By targeting inference over training and focusing on LLMs instead of traditional AI models, HyperAccel is able to overcome cost barriers by designing the most efficient hardware solutions catered to GenAI. With their Bertha-Edge (BerthaE) AI chip designed to enable Generative and Agentic AI at the edge, HyperAccel was honored as one of the winners of the AI Core Technologies & Chips category in the IC Taiwan Grand Challenge (ICTGC) Batch 3.

Finding the Right Balance of Features is Still Necessary for AI Deployment

The issue of cost and hardware requirements has been a constant bottleneck in AI deployment, especially for high-security use cases. Currently, users have to find a balance between performance, versatility, security/ privacy, price, and latency when considering how they want to deploy their AI. For example, applications that need high privacy and security with low latency might have to forego the versatility offered by cloud AI. With their Bertha-Edge (BerthaE) AI chip, HyperAccel is able to streamline memory access to achieve 90% bandwidth utilization compared to 50% commonly seen in GPUs, enabling Generative and Agentic AI at the edge.

Introducing IC Taiwan Grand Challenge

To strengthen Taiwan's position as a global semiconductor powerhouse, the National Science and Technology Council (NSTC) organizes the IC Taiwan Grand Challenge. Started in 2024, the competition originally focused purely on innovative IC designs and advanced application solutions. Today, ICTGC also honors technologies that can support and contribute to the semiconductor industry and Taiwan’s industry in general. ICTGC Batch 3 emphasizes key fields such as AI Core Technologies & Chips, Smart Mobility, Smart Manufacturing, Smart Medtech, and Sustainability.

Empowering a World Where Advanced Technologies are Accessible to All

HyperAccel was founded in 2023 by an associate professor at the Korea Advanced Institute of Science & Technology (KAIST) with research specializing in AI semiconductor. HyperAccel now has exceptional talents from Samsung, as well as PhD and Master's program graduates from KAIST and other prestigious universities. They have won various awards and established a partnership with AMD, Samsung Foundry, and more.

The HyperAccel Bertha-Edge (BerthaE) technology is backed up by 10 Korean patents and 15 patents in the US and Patent Cooperation Treaty (PCT). Accompanied by comprehensive support for deployment, including firmware and software SDKs, the Bertha-Edge (BerthaE) is designed to enable easy integration into business users' edge computing environments. This will allow for rapid deployment of AI-powered applications while reducing the time-to-market for edge AI products.

Designed to efficiently run LLM-based services on the edge, Bertha-Edge (BerthaE) is expected to enable GenAI in on-premise datacenters and help realize the potential of Agentic AI. Current LLM inference is severely limited by memory speed, and Bertha-Edge (BerthaE) has LLM-optimized memory access that maximizes the effective memory bandwidth usage during end-to-end inference regardless of the model configuration. With its streamlined memory access feature, the Bertha-Edge (BerthaE) technology perfectly maintains bandwidth utilization to 90% and removes the need for data reshaping with software-hardware co-design for continuous bandwidth usage. It also features optimized dataflow, compute resources, and chip-to-chip networking for efficient LLM inference. The Bertha-Edge (BerthaE) will allow users to perform data synchronization with low latency and latency hiding to achieve near-perfect scalability. By utilizing LPDDR memory, the Bertha-Edge offers greater yield reliability during production at a lower cost, reducing the overall manufacturing cost and improving production scalability. Overall, Bertha-Edge delivers 3.2x higher performance, 3x better power efficiency, and 10x better cost efficiency than GPUs.

As the winner of the AI Core Technologies & Chips category in the IC Taiwan Grand Challenge Batch 3, HyperAccel will showcase their Bertha-Edge (BerthaE) technology in the Taiwan Innotech Expo (TIE) 2025, which will be held from October 16 - 18, 2025, in the Taipei World Trade Center (TWTC) Hall 1. Visitors are encouraged to join and see how it can accelerate the deployment of AI in various applications.

Register for the IC Taiwan Grand Challenge Batch 4

With the continued success of the IC Taiwan Grand Challenge (ICTGC), the organizers will promote it as a long-term event to support semiconductor innovations from Taiwan and abroad. Registration for the next batch of ICTGC will begin in November 2025, and teams with a focus on semiconductor technologies are encouraged to join and compete.

For more details, please visit:
>>> https://ictaiwanchallenge.org/