Development of Intelligent AI Chips
AI Solutions Fueled by Continuous Innovation
The evolution of intelligent semiconductors has progressed through multiple generations. First-generation CPUs and GPUs, followed by second-generation FPGAs and ASICs, have played a crucial role in advancing AI processing. However, all of these technologies rely on the traditional von Neumann architecture, which requires constant data transfer between the CPU and memory—resulting in high energy consumption. At Pebble Square, we have pioneered Analog-PIM (Processing-In-Memory) AI chips, which revolutionize efficiency by eliminating unnecessary data movement. By integrating computation directly into memory, our PIM technology maximizes performance and drastically reduces power consumption, setting a new standard for AI hardware innovation
-
1st. GEN.
-
2nd. GEN.
-
3rd. GEN.
PIM-based AI Chip Technologies
Efficient, High-Performance AI Processing with Pebble AI Studio
Our PIM-based AI chips enable the deployment of lightweight AI models through our integrated SDK (Pebble AI Studio). Designed for optimal efficiency, our technology minimizes power consumption while delivering high-speed, low-latency performance in a compact form factor
Key Innovations
-
1.
Crossbar Array for Analog Computing – Enhancing AI processing efficiency with advanced analog computation
-
2.
256-State Synaptic Cell (8-bit) – High-precision memory architecture for optimized neural network performance
-
3.
Integrated SDK (Pebble AI Studio) – Seamless AI model deployment and optimization for on-device intelligence
-
4.
Core Advantages – Ultra-Low Power Consumption / Low Latency / Compact
▪ Analog-PIM (Processing-In-Memory)
-
1.Crossbar Arrays for Analog Computing
Our crossbar array architecture enables high-speed, energy-efficient AI processing by directly performing computations in memory. By supporting analog computation, this structure eliminates the need for constant data transfers, significantly reducing power consumption while enhancing processing efficiency
-
2.256-State Synaptic Cell (8-bit)
Leveraging synaptic cells with 8-bit (256 analog states), our technology enables simultaneous processing of complex calculations. This architecture enhances computational efficiency, optimizing performance for advanced AI workloads while maintaining low power consumption
-
3.Integrated SDK (Pebble AI Studio)
To accelerate AI application development, Pebble AI Studio provides an integrated SDK that enables fast and seamless deployment of AI models onto our PIM-based AI chips. This streamlined platform simplifies model optimization, ensuring efficient execution for a wide range of AI applications
Pebble Square’s AI Chip Family
On-Device AI Applications
Unlocking Intelligent, Real-Time Processing Across Industries
-
Sound
A high-performance, ultra-low power AI chip-based speech recognition solution that enables fast and accurate real-time voice recognition. This technology enhances user experience by providing a natural and efficient interface for applications such as smart homes, wearable devices, and mobile devices
-
Vision
An AI-optimized chip designed to efficiently process diverse image data, delivering high-speed, low-power AI vision solutions. Ideal for industries such as healthcare, manufacturing, and beyond, it enhances real-time image analysis and decision-making
-
Security
AI-powered security processing without reliance on internet or network connectivity. With independent on-device data processing, it detects and responds to threats in real time, ensuring a safer, more reliable security solution for various applications
AI-driven Failure Analysis
Enhancing Reliability with Real-Time AI Insights
-
Risk Managements
Our AI-powered solution analyzes real-time data to detect irregularities, identifying abnormal patterns and potential risks in advance. By continuously monitoring system data, it enables proactive risk mitigation and ensures fast, efficient responses to prevent failures before they occur
-
Fault Analysis / Predictive Maintenance
Leveraging AI chip technology, this solution enables real-time diagnostics for automated robots and machinery, detecting potential failures in advance. By identifying issues early, it supports predictive maintenance, minimizing downtime and ensuring stable, uninterrupted operations
