Expedera

Company Overview

Expedera is a provider of scalable neural engine semiconductor intellectual property (IP). The company develops and licenses AI accelerator IP for edge devices, with a focus on delivering high performance and power efficiency for on-device AI inference applications.

Headquartered in Santa Clara, California, Expedera has engineering development centers and customer support offices in the United Kingdom, China, Japan, Taiwan, Singapore, and India. The company’s neural processing unit (NPU) solutions have been successfully deployed in over 10 million consumer devices.

Products Overview

Expedera’s main product line is called Origin, which consists of customizable NPU IP cores optimized for edge AI inference:

• Origin E1: Designed for low-power applications like home appliances and always-on smartphone features. Offers around 1 TOPS of performance.

• Origin E2: Balanced solution for smartphones and edge devices. Offers 1-20 TOPS of performance.

• Origin E6: Higher performance option for next-gen smartphones, AR/VR, and automotive. Offers 16-32 TOPS.

• Origin E8: Highest performance option for automotive/ADAS and data centers. Offers 32-128 TOPS in a single core.

• TimbreAI T3: Ultra-low power AI engine for noise reduction in audio devices.

The Origin NPUs use a unique packet-based architecture that enables parallel execution across multiple neural network layers. This allows for high utilization (up to 90%) and power efficiency (up to 18 TOPS/W). The IP is customizable to specific use cases and supports a wide range of AI model types.

Expedera also provides a full TVM-based software stack to support deployment of trained neural networks on their hardware. This includes tools for importing models, quantization, compilation, profiling, and multi-job APIs.

Founding Team

The founding team of Expedera includes:

• Da Chuang - CEO & Co-founder Previously co-founder and COO of Memoir Systems. Brings over 25 years of ASIC experience from companies like Cisco and Nvidia.

• Siyad Ma - VP Engineering & Co-founder Led ASIC and IP teams at Cisco previously. Over 25 years of experience in ASIC design.

• Sharad Chole - Chief Scientist & Co-founder Expert in AI frameworks and neural network optimizations. Previously an architect at Cisco and Microsoft.

Problem and Market Fit

Expedera is addressing the growing need for efficient AI processing at the edge. As more AI capabilities are pushed to edge devices like smartphones, IoT devices, and vehicles, there is increasing demand for NPU solutions that can deliver high performance within tight power and area constraints.

The company’s customizable IP allows chip designers to optimize AI acceleration for specific use cases and performance targets. This helps device makers differentiate their products with enhanced on-device AI capabilities while managing power consumption.

Business Model

Expedera operates on a semiconductor IP licensing model. They develop and license their NPU designs to chipmakers and OEMs, who can then integrate the IP into their system-on-chip (SoC) designs. This allows Expedera’s customers to add AI acceleration capabilities to their chips without having to develop the NPU technology in-house.

The company likely generates revenue through upfront licensing fees as well as per-unit royalties on chips shipped with their IP. They may also provide engineering services to help customers integrate and optimize the IP for their specific applications.

Competitive Landscape

The AI accelerator IP market has several established and emerging players, including:

• Arm - Offers Mali NPU IP
• Imagination Technologies - PowerVR Neural Network Accelerator
• Cadence - Tensilica AI accelerators
• Synopsys - DesignWare ARC NPX processors
• Ceva - NeuPro AI processors

Expedera aims to differentiate itself through its highly efficient packet-based architecture, customizability, and strong power/performance metrics. The company claims its NPUs can achieve higher sustained utilization rates compared to competitors.

Customers

While Expedera does not publicly disclose its full customer list, the company states that its IP has been deployed in over 10 million consumer devices. Based on the applications highlighted, likely customers include:

• Smartphone manufacturers
• AR/VR headset makers
• Automotive/ADAS system developers
• IoT/smart home device makers
• Consumer electronics companies

Relevant News

• January 2024 - Expedera announced its Origin E6 NPU IP, targeting next-generation smartphones and AR/VR devices with 16-32 TOPS of AI performance.

• June 2023 - The company introduced its packet-based AI processing architecture, highlighting its advantages over traditional layer-based NPU designs.

• November 2022 - Expedera raised $30M in Series B funding led by Industrifonden to accelerate product development and expand its global footprint.

This comprehensive overview provides key details about Expedera’s technology, products, team, and market position that would be relevant to potential investors or partners interested in the edge AI acceleration space. The company appears well-positioned to capitalize on the growing demand for efficient on-device AI processing capabilities across multiple industries.