All IPs > Processor > AI Processor
The AI Processor category within our semiconductor IP catalog is dedicated to state-of-the-art technologies that empower artificial intelligence applications across various industries. AI processors are specialized computing engines designed to accelerate machine learning tasks and perform complex algorithms efficiently. This category includes a diverse collection of semiconductor IPs that are built to enhance both performance and power efficiency in AI-driven devices.
AI processors play a critical role in the emerging world of AI and machine learning, where fast processing of vast datasets is crucial. These processors can be found in a range of applications from consumer electronics like smartphones and smart home devices to advanced robotics and autonomous vehicles. By facilitating rapid computations necessary for AI tasks such as neural network training and inference, these IP cores enable smarter, more responsive, and capable systems.
In this category, developers and designers will find semiconductor IPs that provide various levels of processing power and architectural designs to suit different AI applications, including neural processing units (NPUs), tensor processing units (TPUs), and other AI accelerators. The availability of such highly specialized IPs ensures that developers can integrate AI functionalities into their products swiftly and efficiently, reducing development time and costs.
As AI technology continues to evolve, the demand for robust and scalable AI processors increases. Our semiconductor IP offerings in this category are designed to meet the challenges of rapidly advancing AI technologies, ensuring that products are future-ready and equipped to handle the complexities of tomorrow’s intelligence-driven tasks. Explore this category to find cutting-edge solutions that drive innovation in artificial intelligence systems today.
The Akida Neural Processor is a sophisticated AI processing unit designed to handle complex neural network tasks with unmatched precision and efficiency. Utilizing an event-based processing model, Akida exploits data sparsity to minimize operations and hence decrease power usage significantly while enhancing throughput. This processor is built around a mesh network interconnect, with each node equipped with configurable Neural Network Engines that can handle convolutional and fully connected neural networks. With these capabilities, Akida can process data at the edge, maintaining high-speed, low-latency responses ideal for real-time applications. Akida maintains seamless functionality in diverse use cases, from predictive maintenance to streaming analytics in sensors. By supporting on-chip learning and providing strong privacy controls, this processor ensures data security by reducing cloud data exchanges, making it a trusted component for sensitive applications.
The 2nd Generation Akida processor introduces groundbreaking enhancements to BrainChip's neuromorphic processing platform, particularly ideal for intricate network models. It integrates eight-bit weight and activation support, improving energy efficiency and computational performance without enlarging model size. By supporting an extensive application set, Akida 2nd Generation addresses diverse Edge AI needs untethered from cloud dependencies. Notably, Akida 2nd Generation incorporates Temporal Event-Based Neural Nets (TENNs) and Vision Transformers, facilitating robust tracking through high-speed vision and audio processing. Its built-in support for on-chip learning further optimizes AI efficiency by reducing reliance on cloud training. This versatile processor fits perfectly for spatio-temporal applications across industrial, automotive, and healthcare sectors. Developers gain from its Configurable IP Platform, which allows seamless scalability across multiple use cases. The Akida ecosystem, including MetaTF, offers developers a strong foundation for integrating cutting-edge AI capabilities into Edge systems, ensuring secure and private data processing.
The NMP-750 is designed as a cutting-edge performance accelerator for edge computing, tailored to address challenges in sectors like automotive, telecommunications, and smart factories. This product offers ample support for mobility, autonomous control, and process automation, setting a benchmark in high-performance computing for varied applications. With a processing power of up to 16 TOPS and 16 MB of local memory, it supports RISC-V/Arm Cortex-R or A 32-bit CPUs for substantial computational tasks. Its architecture supports a rich set of applications, including multi-camera stream processing and energy management, enabled through its AXI4 128-bit interfaces that manage extensive data traffic efficiently. This accelerator is particularly suited for complex scenarios such as spectral efficiency and smart building management, offering unparalleled performance capabilities. Designed for scalability and reliability, the NMP-750 reaches beyond traditional computing barriers, ensuring outstanding performance in real-time applications and next-gen technology deployments.
The KL730 AI SoC is equipped with a state-of-the-art third-generation reconfigurable NPU architecture, delivering up to 8 TOPS of computational power. This innovative architecture enhances computational efficiency, particularly with the latest CNN networks and transformer applications, while reducing DDR bandwidth demands. The KL730 excels in video processing, offering support for 4K 60FPS output and boasts capabilities like noise reduction, wide dynamic range, and low-light imaging. It is ideal for applications such as intelligent security, autonomous driving, and video conferencing.
The NMP-350 is specifically designed to serve as a cost-effective endpoint accelerator with a strong emphasis on low power consumption, making it ideal for various applications in AIoT, automotive, and smart appliances. This product is equipped with a robust architecture to facilitate myriad applications, such as driver authentication, digital mirrors, and predictive maintenance, while ensuring efficient resource management. Capable of delivering up to 1 TOPS, the NMP-350 integrates up to 1 MB of local memory, supporting RISC-V/Arm Cortex-M 32-bit CPU cores. It utilizes a triple AXI4 interface, each with a capacity of 128 bits, to manage host, CPU, and data traffic seamlessly. This architecture supports a host of applications in wearables, Industry 4.0, and health monitoring, adding significant value to futuristic technology solutions. Strategically targeting markets like AIoT/sensors and smart appliances, the NMP-350 positions itself as a favored choice for developing low-cost, power-sensitive device solutions. As industries gravitate toward energy-efficient technologies, products like NMP-350 offer a competitive edge in facilitating smart, green development processes.
Designed for high-performance applications, the Metis AIPU PCIe AI Accelerator Card employs four Metis AI Processing Units to deliver exceptional computational power. With its ability to reach up to 856 TOPS, this card is tailored for demanding vision applications, making it suitable for real-time processing of multi-channel video data. The PCIe form factor ensures easy integration into existing systems, while the customized software platform simplifies the deployment of neural networks for tasks like YOLO object detection. This accelerator card ensures scalability and efficiency, allowing developers to implement AI applications that are both powerful and cost-effective. The card’s architecture also takes advantage of RISC-V and Digital-In-Memory Computing technologies, bringing substantial improvements in speed and power efficiency.
The Origin E1 neural engines by Expedera redefine efficiency and customization for low-power AI solutions. Specially crafted for edge devices like home appliances and security cameras, these engines serve ultra-low power applications that demand continuous sensing capabilities. They minimize power consumption to as low as 10-20mW, keeping data secure and eliminating the need for external memory access. The advanced packet-based architecture enhances performance by facilitating parallel layer execution, thereby optimizing resource utilization. Designed to be a perfect fit for dedicated AI functions, Origin E1 is tailored to support specific neural networks efficiently while reducing silicon area and system costs. It supports various neural networks, from CNNs to RNNs, making it versatile for numerous applications. This engine is also one of the most power-efficient in the industry, boasting an impressive 18 TOPS per Watt. Origin E1 also offers a full TVM-based software stack for easy integration and performance optimization across customer platforms. It supports a wide array of data types and networks, ensuring flexibility and sustained power efficiency, averaging 80% utilization. This makes it a reliable choice for OEMs looking for high performance in always-sensing applications, offering a competitive edge in both power efficiency and security.
The Veyron V2 CPU takes the innovation witnessed in its predecessor and propels it further, offering unparalleled performance for AI and data center-class applications. This successor to the V1 CPU integrates seamlessly into environments requiring high computational power and efficiency, making it perfect for modern data challenges. Built upon RISC-V's architecture, it provides an open-standard alternative to traditional closed processor models. With a heavy emphasis on AI and machine learning workloads, Veyron V2 is designed to excel in handling complex data-centric tasks. This CPU can quickly adapt to multifaceted requirements, proving indispensable from enterprise servers to hyperscale data centers. Its superior design enables it to outperform many contemporary alternatives, positioning it as a lead component for next-generation computing solutions. The processor's adaptability allows for rapid and smooth integration into existing systems, facilitating quick upgrades and enhancements tailored to specific operational needs. As the Veyron V2 CPU is highly energy-efficient, it empowers data centers to achieve greater sustainability benchmarks without sacrificing performance.
The Metis AIPU M.2 Accelerator Module is a cutting-edge AI processing unit designed to boost the performance of edge computing tasks. This module integrates seamlessly with innovative applications, offering a robust solution for inference at the edge. It excels in vision AI tasks with its dedicated 512MB LPDDR4x memory, providing the necessary storage for complex tasks. Offering unmatched energy efficiency, the Metis AIPU M.2 module is capable of delivering significant performance gains while maintaining minimal power consumption. At an accessible price point, this module opens up AI processing capabilities for a variety of applications. As an essential component of next-generation vision processing systems, it is ideal for industries seeking to implement AI technologies swiftly and effectively.
The Origin E8 NPUs represent Expedera's cutting-edge solution for environments demanding the utmost in processing power and efficiency. This high-performance core scales its TOPS capacity between 32 and 128 with single-core configurations, addressing complex AI tasks in automotive and data-centric operational settings. The E8’s architecture stands apart due to its capability to handle multiple concurrent tasks without any compromise in performance. This unit adopts Expedera's signature packet-based architecture for optimized parallel execution and resource management, removing the necessity for hardware-specific tweaks. The Origin E8 also supports high input resolutions up to 8K and integrates well across standard and custom neural networks, enhancing its utility in future-forward AI applications. Leveraging a flexible, scalable design, the E8 IP cores make use of an exhaustive software suite to augment AI deployment. Field-proven and already deployed in a multitude of consumer vehicles, Expedera's Origin E8 provides a robust, reliable choice for developers needing optimized AI inference performance, ideally suited for data centers and high-power automobile systems.
The SCR9 Processor Core is a cutting-edge processor designed for entry-level server-class and personal computing applications. Featuring a 12-stage dual-issue out-of-order pipeline, it supports robust RISC-V extensions including vector operations and a high-complexity memory system. This core is well-suited for high-performance computing, offering exceptional power efficiency with multicore coherence and the ability to integrate accelerators, making it suitable for areas like AI, ML, and enterprise computing.
Akida IP stands as an advanced neuromorphic processor, emulating brain-like processing to efficiently handle sensor inputs at acquisition points. This digital processor offers superior performance, precision, and significant reductions in power usage. By facilitating localized AI/ML tasks, it decreases latency and enhances data privacy. Akida IP is built to infer and learn at the edge, offering highly customizable, event-based neural processing. The architecture of Akida IP is scalable and compact, supporting an extensive mesh network connection of up to 256 nodes. Each node includes four Neural Network Layer Engines (NPEs), configurable for convolutional and fully connected processes. By leveraging data sparsity, Akida optimizes operation reduction, making it a cost-effective solution for various edge AI applications. Including MetaTF support for model simulations, Akida IP brings a fully synthesizable RTL IP package compatible with standard EDA tools, emphasizing ease of integration and deployment. This enables developers to swiftly design, develop, and implement custom AI solutions with robust security and privacy protection.
The NaviSoC, a flagship product of ChipCraft, combines a GNSS receiver with an on-chip application processor, providing an all-in-one solution for high-precision navigation and timing applications. This product is designed to meet the rigorous demands of industries such as automotive, UAVs, and smart agriculture. One of its standout features is the ability to support all major global navigation satellite systems, offering versatile functionality for various professional uses. The NaviSoC is tailored for high efficiency, delivering performance that incorporates low power consumption with robust computational capabilities. Specifically tailored for next-generation applications, NaviSoC offers flexibility through its ability to be adapted for different tasks, making it a preferred choice for many industries. It integrates seamlessly into systems requiring precision and reliability, providing developers with a wide array of programmable peripherals and interfaces. The foundational design ethos of the NaviSoC revolves around minimizing power usage while ensuring high precision and accuracy, making it an ideal component for battery-powered and portable devices. Additionally, ChipCraft provides integrated software development tools and navigation firmware, ensuring that clients can capitalize on fast time-to-market for their products. The design of the NaviSoC takes a comprehensive approach, factoring in real-world application requirements such as temperature variation and environmental challenges, thus providing a resilient and adaptable product for diverse uses.
The High Performance RISC-V Processor from Cortus represents the forefront of high-end computing, designed for applications demanding exceptional processing speeds and throughput. It features an out-of-order execution core that supports both single-core and multi-core configurations for diverse computing environments. This processor specializes in handling complex tasks requiring multi-threading and cache coherency, making it suitable for applications ranging from desktops and laptops to high-end servers and supercomputers. It includes integrated vector and AI accelerators, enhancing its capability to manage intensive data-processing workloads efficiently. Furthermore, this RISC-V processor is adaptable for advanced embedded systems, including automotive central units and AI applications in ADAS, providing enormous potential for innovation and performance across various markets.
Origin E2 NPUs focus on delivering power and area efficiency, making them ideal for on-device AI applications in smartphones and edge nodes. These processing units support a wide range of neural networks, including video, audio, and text-based applications, all while maintaining impressive performance metrics. The unique packet-based architecture ensures effective performance with minimal latency and eliminates the need for hardware-specific optimizations. The E2 series offers customization options allowing it to fit specific application needs perfectly, with configurations supporting up to 20 TOPS. This flexibility represents significant design advancements that help increase processing efficiency without introducing latency penalties. Expedera's power-efficient design results in NPUs with industry-leading performance at 18 TOPS per Watt. Further augmenting the value of E2 NPUs is their ability to run multiple neural network types efficiently, including LLMs, CNNs, RNNs, and others. The IP is field-proven, deployed in over 10 million consumer devices, reinforcing its reliability and effectiveness in real-world applications. This makes the Origin E2 an excellent choice for companies aiming to enhance AI capabilities while managing power and area constraints effectively.
NeuroMosAIc Studio is a comprehensive software platform designed to support AI modeling and deployment, offering a suite of tools that streamline the development of neural networks for high-performance computing tasks. This solution facilitates converting, compressing, and optimizing AI models ready for integration with AiM Future’s hardware accelerators. The platform includes features like network quantization, using formats like 1-bit, FXP8, and FXP16, and network optimization for efficient hardware generation. Advanced mapping tools and precision analysis components are in place to ensure optimal deployment and performance alignment with AiM Future's accelerators. Users can leverage the studio's compilation, simulation, and AI-aware training capabilities to refine AI models both in the cloud and at the edge, optimizing quantization and adjustment processes. NeuroMosAIc Studio is pivotal in enhancing the performance of AI solutions across various applications, from smart city management to advanced AR/VR experiences.
Cortus's Automotive AI Inference SoC is a breakthrough solution tailored for autonomous driving and advanced driver assistance systems. This SoC combines efficient image processing with AI inference capabilities, optimized for city infrastructure and mid-range vehicle markets. Built on a RISC-V architecture, the AI Inference SoC is capable of running specialized algorithms, akin to those in the Yolo series, for fast and accurate image recognition. Its low power consumption makes it suitable for embedded automotive applications requiring enhanced processing without compromising energy efficiency. This chip demonstrates its adequacy for Level 2 and Level 4 autonomous driving systems, providing a comprehensive AI-driven platform that enhances safety and operational capabilities in urban settings.
The Chimera GPNPU by Quadric is a versatile processor specifically designed to enhance machine learning inference tasks on a broad range of devices. It provides a seamless blend of traditional digital signal processing (DSP) and neural processing unit (NPU) capabilities, which allow it to handle complex ML networks alongside conventional C++ code. Designed with a focus on adaptability, the Chimera GPNPU architecture enables easy porting of various models and software application programming, making it a robust solution for rapidly evolving AI technologies. A key feature of the Chimera GPNPU is its scalable design, which extends from 1 to a remarkable 864 TOPs, catering to applications from standard to advanced high-performance requirements. This scalability is coupled with its ability to support a broad range of ML networks, such as classic backbones, vision transformers, and large language models, fulfilling various computational needs across industries. The Chimera GPNPU also excels in automotive applications, including ADAS and ECU systems, due to its ASIL-ready design. The processor's hybrid architecture merges Von Neumann and 2D SIMD matrix capabilities, promoting efficient execution of scalar, vector, and matrix operations. It boasts a deterministic execution pipeline and extensive customization options, including configurable instruction caches and local register memories that optimize memory usage and power efficiency. This design effectively reduces off-chip memory accesses, ensuring high performance while minimizing power consumption.
The KL630 AI SoC embodies next-generation AI chip technology with a pioneering NPU architecture. It uniquely supports Int4 precision and transformer networks, offering superb computational efficiency combined with low power consumption. Utilizing an ARM Cortex A5 CPU, it supports a range of AI frameworks and is built to handle scenarios from smart security to automotives, providing robust capability in both high and low light conditions.
The NMP-550 stands out as a performance-focused accelerator catered towards applications necessitating high efficiency, especially in demanding fields such as automotive, drones, and AR/VR. This technology caters to various application needs, including driver monitoring, image/video analytics, and heightened security measures through its powerful architecture and processing capability. Boasting a significant computation potential of up to 6 TOPS, the NMP-550 includes up to 6 MB of local memory. Featuring RISC-V/Arm Cortex-M or A 32-bit CPUs, the product ensures robust processing for advanced applications. The triple AXI4 interface provides a seamless 128-bit data exchange across hosts, CPUs, and data channels, magnifying flexibility for technology integrators. Ideal for medical devices, this product also expands its utility into security and surveillance, supporting crucial processes like super-resolution and fleet management. Its comprehensive design and efficiency make it an optimal choice for applications demanding elevated performance within constrained resources.
The Origin E6 neural engines are built to push the boundaries of what's possible in edge AI applications. Supporting the latest in AI model innovations, such as generative AI and various traditional networks, the E6 scales from 16 to 32 TOPS, aimed at balancing performance, efficiency, and flexibility. This versatility is essential for high-demand applications in next-generation devices like smartphones, digital reality setups, and consumer electronics. Expedera’s E6 employs packet-based architecture, facilitating parallel execution that leads to optimal resource usage and eliminating the need for dedicated hardware optimizations. A standout feature of this IP is its ability to maintain up to 90% processor utilization even in complex multi-network environments, thus proving its robustness and adaptability. Crafted to fit various use cases precisely, E6 offers a comprehensive TVM-based software stack and is well-suited for tasks that require simultaneous running of numerous neural networks. This has been proven through its deployment in over 10 million consumer units. Its design effectively manages power and system resources, thus minimizing latency and maximizing throughput in demanding scenarios.
The xcore.ai platform by XMOS Semiconductor is a sophisticated and cost-effective solution aimed specifically at intelligent IoT applications. Harnessing a unique multi-threaded micro-architecture, xcore.ai provides superior low latency and highly predictable performance, tailored for diverse industrial needs. It is equipped with 16 logical cores divided across two multi-threaded processing tiles. These tiles come enhanced with 512 kB of SRAM and a vector unit supporting both integer and floating-point operations, allowing it to process both simple and complex computational demands efficiently. A key feature of the xcore.ai platform is its powerful interprocessor communication infrastructure, which enables seamless high-speed communication between processors, facilitating ultimate scalability across multiple systems on a chip. Within this homogeneous environment, developers can comfortably integrate DSP, AI/ML, control, and I/O functionalities, allowing the device to adapt to specific application requirements efficiently. Moreover, the software-defined architecture allows optimal configuration, reducing power consumption and achieving cost-effective intelligent solutions. The xcore.ai platform shows impressive DSP capabilities, thanks to its scalar pipeline that achieves up to 32-bit floating-point operations and peak performance rates of up to 1600 MFLOPS. AI/ML capabilities are also robust, with support for various bit vector operations, making the platform a strong contender for AI applications requiring homogeneous computing environments and exceptional operator integration.
The Titanium Ti375 FPGA from Efinix boasts a high-density, low-power configuration, ideal for numerous advanced computing applications. Built on the well-regarded Quantum compute fabric, this FPGA integrates a robust set of features including a hardened RISC-V block, SerDes transceiver, and LPDDR4 DRAM controller, enhancing its versatility in challenging environments. The Ti375 model is designed with an intuitive I/O interface, allowing seamless communication and data handling. Its innovative architecture ensures minimal power consumption without compromising on processing speed, making it highly suitable for portable and edge devices. The inclusion of MIPI D-PHY further expands its applications in image processing and high-speed data transmission tasks. This FPGA is aligned with current market demands, emphasizing efficiency and scalability. Its architecture allows for diverse design challenges, supporting applications that transcend traditional boundaries. Efinix’s commitment to delivering sophisticated yet energy-efficient solutions is embodied in the Titanium Ti375, enabling new possibilities in the realm of computing.
The Matchstiq™ X40 by Epiq Solutions is a compact, high-performance software-defined radio (SDR) system designed to harness the power of AI and machine learning at the RF edge. Its small form factor makes it suitable for payloads with size, weight, and power constraints. The unit offers RF coverage up to 18GHz with an instantaneous bandwidth up to 450MHz, making it an excellent choice for demanding environments requiring advanced signal processing and direction finding. One of the standout features of the Matchstiq™ X40 is its integration of Nvidia's Orin NX for CPU/GPU operations and an AMD Zynq Ultrascale+ FPGA, allowing for sophisticated data processing capabilities directly at the point of RF capture. This combination offers enhanced performance for real-time signal analysis and machine learning implementations, making it suited for a variety of high-tech applications. The device supports a variety of input/output configurations, including 1 GbE, USB 3.0, and GPSDO, ensuring compatibility with numerous host systems. It offers dual configurations that support up to four receivers and two transmitters, along with options for phase-coherent multi-channel operations, thereby broadening its usability across different mission-critical tasks.
The RWM6050 baseband modem by Blu Wireless represents a highly efficient advancement in mmWave technology, offering an economical and energy-saving option for high bandwidth and capacity applications. Developed alongside Renesas, the modem is configured to work with mmWave RF chipsets to deliver scalable multi-gigabit throughput across access and backhaul networks. This modem is ideal for applications requiring substantial data transfer across several hundred meters.\n\nThe RWM6050 leverages flexible channelization and advanced modulation support to enhance data rates with dual modems and integrated mixed-signal front-end processing. This ensures that the modem can effectively handle diverse use cases with varying bandwidth demands. Its versatile subsystems, including PHY, MAC, ADC/DAC, and beamforming, facilitate adaptive solutions for complex networking environments.\n\nA standout feature of the RWM6050 is its integrated network synchronization, ensuring high precision in data delivery. Designed to meet the futuristic needs of communication networks, it helps end-users achieve superior performance through its programmable real-time scheduler and digital front-end processing. Additionally, the modem's highly digital design supports robust, secure connections needed for next-generation connectivity solutions.
The Ultra-Low-Power 64-Bit RISC-V Core by Micro Magic represents a significant advancement in energy-efficient computing. This core, operating at an astonishingly low 10mW while running at 1GHz, sets a new standard for low-power design in processors. Micro Magic's proprietary methods ensure that this core maintains high performance even at reduced voltages, making it a perfect fit for applications where power conservation is crucial. Micro Magic's RISC-V core is designed to deliver substantial computational power without the typical energy costs associated with traditional architectures. With capabilities that make it suitable for a wide array of high-demand tasks, this core leverages sophisticated design approaches to achieve unprecedented power efficiency. The core's impressive performance metrics are complemented by Micro Magic's specialized tools, which aid in integrating the core into larger systems. Whether for embedded applications or more demanding computational roles, the Ultra-Low-Power 64-Bit RISC-V Core offers a compelling combination of power and performance. The design's flexibility and power efficiency make it a standout among other processors, reaffirming Micro Magic's position as a leader in semiconductor innovation. This solution is poised to influence how future processors balance speed and energy usage significantly.
The Veyron V1 CPU represents an efficient, high-performance processor tailored to address a myriad of data center demands. As an advanced RISC-V architecture processor, it stands out by offering competitive performance compatible with the most current data center workloads. Designed to excel in efficiency, it marries performance with a sustainable energy profile, allowing for optimal deployment in various demanding environments. This processor brings flexibility to developers and data center operators by providing extensive customization options. Veyron V1's robust architecture is meant to enhance throughput and streamline operations, facilitating superior service provision across cloud infrastructures. Its compatibility with diverse integration requirements makes it ideal for a broad swath of industrial uses, encouraging scalability and robust data throughput. Adaptability is a key feature of Veyron V1 CPU, making it a preferred choice for enterprises looking to leverage RISC-V's open standards and extend the performance of their platforms. It aligns seamlessly with Ventana's broader ecosystem of products, creating excellence in workload delivery and resource management within hyperscale and enterprise environments.
The Dynamic Neural Accelerator II by EdgeCortix is a pioneering neural network core that combines flexibility and efficiency to support a broad array of edge AI applications. Engineered with run-time reconfigurable interconnects, it facilitates exceptional parallelism and efficient data handling. The architecture supports both convolutional and transformer neural networks, offering optimal performance across varied AI use cases. This architecture vastly improves upon traditional IP cores by dynamically reconfiguring data paths, which significantly enhances parallel task execution and reduces memory bandwidth usage. By adopting this approach, the DNA-II boosts its processing capability while minimizing energy consumption, making it highly effective for edge AI applications that require high output with minimal power input. Furthermore, the DNA-II's adaptability enables it to tackle inefficiencies often seen in batching tasks across other IP ecosystems. The architecture ensures that high utilization and low power consumption are maintained across operations, profoundly impacting sectors relying on edge AI for real-time data processing and decision-making.
The Trion FPGA family by Efinix addresses the dynamic needs of edge computing and IoT applications. These devices range from 4K to 120K logic elements, balancing computational capability with efficient power usage for a wide range of general-purpose applications. Trion FPGAs are designed to empower edge devices with rapid processing capabilities and flexible interfacing. They support a diverse array of use-cases, from industrial automation systems to consumable electronics requiring enhanced connectivity and real-time data processing. Offering a pragmatic solution for designers, Trion FPGAs integrate seamlessly into existing systems, facilitating swift development and deployment. They provide unparalleled adaptability to meet the intricate demands of modern technological environments, thereby enabling innovative edge and IoT solutions to flourish.
Efinix's Topaz FPGA series is engineered for mass-market applications, delivering a perfect mix of efficiency and adaptability. These FPGAs encapsulate a highly efficient architecture, combined with the industry's essential features and protocols, such as PCIe Gen3, MIPI, and LPDDR4. This configuration allows users to harness substantial performance while maintaining ample room for future innovations. Topaz FPGAs are optimized for high-volume production environments where cost-effectiveness and swift integration are paramount. Their design promotes ease of implementation in various applications, spanning from automotive to deeply embedded systems, where reliability and robustness are key. Featuring a streamlined architecture, Topaz series FPGAs support modern connectivity standards and data processing capabilities. These devices are tailored for industries requiring scalable solutions that can adapt to evolving technological landscapes, ensuring that Efinix customers remain competitive in their respective fields.
The AON1020 expands AI processing capabilities to encompass not only voice and audio recognition but also a variety of sensor applications. It leverages the power of the AONSens Neural Network cores, offering a comprehensive solution that integrates Verilog RTL technology to support both ASIC and FPGA products. Key to the AON1020's appeal is its versatility in addressing various sensor data, such as human activity detection. This makes it indispensable in applications requiring nuanced responses to environmental inputs, from motion to gesture awareness. It deploys these capabilities while minimizing energy demands, aligning perfectly with the needs of battery-operated and wearable devices. By executing real-time analytics on device-stored data, the AON1020 ensures high accuracy in environments fraught with noise and user variability. Its architecture allows it to detect multiple commands simultaneously, enhancing device interaction while maintaining low power consumption. Thus, the AON1020 is not only an innovator in sensor data interaction but also a leader in ensuring extended device functionality without compromising energy efficiency or processing accuracy.
VSORA's Tyr Superchip epitomizes high-performance capabilities tailored for the demanding worlds of autonomous driving and generative AI. With its advanced multi-core architecture, this superchip can execute any algorithm efficiently without relying on CUDA, which promotes versatility in AI deployment. Built to deliver a seamless combination of AI and general-purpose processing, the Tyr Superchip utilizes sparsity techniques, supporting quantization on-the-fly, which optimizes its performance for a wide array of computational tasks. The Tyr Superchip is distinctive for its ability to support the simultaneous execution of AI and DSP tasks, selectable on a layer-by-layer basis, which provides unparalleled flexibility in workload management. This flexibility is further complemented by its low latency and power-efficient design, boasting performance near theoretical maximums, with support for next-generation algorithms and software-defined vehicles (SDVs). Safety is prioritized with the implementation of ISO26262/ASIL-D features, making the Tyr Superchip an ideal solution for the automotive industry. Its hardware is designed to handle the computational load required for safe and efficient autonomous driving, and its programmability allows for ongoing adaptations to new automotive standards and innovations.
The Low Power RISC-V CPU IP from SkyeChip is crafted to deliver efficient computation with minimal power consumption. Featuring the RISC-V RV32 instruction set, it supports a range of functions with full standard compliance for instruction sets and partial support where necessary. Designed exclusively for machine mode, it incorporates multiple vectorized interrupts and includes comprehensive debugging capabilities. This CPU IP is well-suited for integration into embedded systems where power efficiency and processing capability are crucial.
The NPU, part of the ENLIGHT series by OPENEDGES Technology, is designed as a deep learning accelerator focusing on inferencing computations with superior efficiency and compute density. Developed for high-performance edge computing, this neural processing unit supports a range of operations pertinent to deep neural networks, including convolution and pooling, providing state-of-the-art capability in both power and performance. The NPU's architecture is based on mixed-precision computation using 4-/8-bit quantization which significantly reduces DRAM traffic, thereby optimizing bandwidth utilization and power consumption. Its design incorporates an advanced vector engine optimized for modern deep neural network architectures, enriching its ability to modernize and scale with evolving AI workloads. Accompanying the hardware capabilities, the NPU offers a comprehensive software toolkit featuring network conversion, quantization, and simulation tools. This suite is built for compatibility with mainstream AI frameworks and ensures seamless integration and efficiency in real-world applications ranging from automotive systems to surveillance.
The 100 Gbps Polar Encoder and Decoder is engineered for the next-generation communication systems demanding ultra-high data rates and reliability. It employs Polar coding, a recent advancement in code theory, which provides a capacity achieving solution to enhance data transfer efficiency in modern networks, particularly suitable for 5G technologies. This IP core supports data rates up to 100 Gbps, enabling rapid data encoding and decoding essential for high-speed communication backbones. The technology ensures robust error correction and maximal utilization of spectral resources by leveraging the power of Polar code combined with optimized algorithmic implementations. Strategically designed for industry-leading performance, this Polar Encoder and Decoder is applicable in systems where bandwidth efficiency and processing speed are critical. It is highly applicable to the telecommunication industries involved in mobile networks, data centers, and any large-scale data streaming operations.
Avispado is a sophisticated 64-bit RISC-V core that emphasizes efficiency and adaptability within in-order execution frameworks. It's engineered to cater to energy-efficient SoC designs, making it an excellent choice for machine learning applications with its compact design and ability to seamlessly communicate with RISC-V Vector Units. By utilizing the Gazzillion Misses™ technology, the Avispado core effectively handles high sparsity in tensor weights, resulting in superior energy efficiency per operation. This core features a 2-wide in-order configuration and supports the RISC-V Vector Specification 1.0 as well as Semidynamics' Open Vector Interface. With support for large memory capacities, it includes complete MMU features and is Linux-ready, ensuring it's prepared for demanding computational tasks. The core's native CHI interface can be fine-tuned to AXI, promoting cache-coherent multiprocessing capabilities. Avispado is optimized for various demanding workloads, with optional extensions for specific needs such as bit manipulation and cryptography. The core's customizable configuration allows changes to its instruction and data cache sizes (I$ and D$ from 8KB to 32KB), ensuring it meets specific application demands while retaining operational efficiency.
The Cortus ULYSS range of automotive microcontrollers is engineered to meet the demands of sophisticated automotive applications, extending from body control to ADAS and infotainment systems. Utilizing a RISC-V architecture, these microcontrollers provide high performance and efficiency suitable for automotive tasks. Each variant within the ULYSS family caters to specific automotive functions, with capabilities ranging from basic energy management to complex networking and ADAS processing. For instance, the ULYSS1 caters to body control applications with a single-core CPU, while the ULYSS3 provides robust networking capabilities with a quad-core, lockstep MPU operating up to 1.5 GHz. The ULYSS line is structured to offer scalability and flexibility, allowing automotive manufacturers to integrate these solutions seamlessly into various components of a vehicle's electronic system. This focus on adaptability helps Cortus provide both a cost-effective and high-performance solution for its automotive partners.
RaiderChip's GenAI v1 is a pioneering hardware-based generative AI accelerator, designed to perform local inference at the Edge. This technology integrates optimally with on-premises servers and embedded devices, offering substantial benefits in privacy, performance, and energy efficiency over traditional hybrid AI solutions. The design of the GenAI v1 NPU streamlines the process of executing large language models by embedding them directly onto the hardware, eliminating the need for external components like CPUs or internet connections. With its ability to support complex models such as the Llama 3.2 with 4-bit quantization on LPDDR4 memory, the GenAI v1 achieves unprecedented efficiency in AI token processing, coupled with energy savings and reduced latency. What sets GenAI v1 apart is its scalability and cost-effectiveness, significantly outperforming competitive solutions such as Intel Gaudi 2, Nvidia's cloud GPUs, and Google's cloud TPUs in terms of memory efficiency. This solution maximizes the number of tokens generated per unit of memory bandwidth, thus addressing one of the primary limitations in generative AI workflow. Furthermore, the adept memory usage of GenAI v1 reduces the dependency on costly memory types like HBM, opening the door to more affordable alternatives without diminishing processing capabilities. With a target-agnostic approach, RaiderChip ensures the GenAI v1 can be adapted to various FPGAs and ASICs, offering configuration flexibility that allows users to balance performance with hardware costs. Its compatibility with a wide range of transformers-based models, including proprietary modifications, ensures GenAI v1's robust placement across sectors requiring high-speed processing, like finance, medical diagnostics, and autonomous systems. RaiderChip's innovation with GenAI v1 focuses on supporting both vanilla and quantized AI models, ensuring high computation speeds necessary for real-time applications without compromising accuracy. This capability underpins their strategic vision of enabling versatile and sustainable AI solutions across industries. By prioritizing integration ease and operational independence, RaiderChip provides a tangible edge in applying generative AI effectively and widely.
Optimized for high-performance tasks, the SCR7 Application Core is a 64-bit RISC-V processor with robust Linux capability. Tailored for powerful data-intensive applications, this core features a 12-stage out-of-order pipeline and supports vector operations, making it ideal for AI, ML, and high-performance computing applications. It integrates seamlessly with multicore environments, offering comprehensive memory management and high-level interrupt systems, facilitated by standard interfaces for broad compatibility.
The EW6181 is a cutting-edge multi-GNSS silicon solution offering the lowest power consumption and high sensitivity for exemplary accuracy across a myriad of navigation applications. This GNSS chip is adept at processing signals from numerous satellite systems including GPS L1, Glonass, BeiDou, Galileo, and several augmentation systems like SBAS. The integrated chip comprises an RF frontend, a digital baseband processor, and an ARM microcontroller dedicated to operating the firmware, allowing for flexible integration across devices needing efficient power usage. Designed with a built-in DC-DC converter and LDOs, the EW6181 silicon streamlines its bill of materials, making it perfect for battery-powered devices, providing extended operational life without compromising on performance. By incorporating patent-protected algorithms, the EW6181 achieves a remarkably compact footprint while delivering superior performance characteristics. Especially suited for dynamic applications such as action cameras and wearables, its antenna diversity capabilities ensure exceptional connectivity and positioning fidelity. Moreover, by enabling cloud functionality, the EW6181 pushes boundaries in power efficiency and accuracy, catering to connected environments where greater precision is paramount.
Hanguang 800 is a sophisticated AI acceleration chip tailored for demanding neural network tasks. Developed with T-Head's cutting-edge technology, it excels in delivering high throughput for deep learning workloads. This chip employs a robust architecture optimized for AI computations, providing unprecedented performance improvements in neural network execution. It's particularly suited for scenarios requiring large-scale AI processing, such as image recognition and natural language processing. The chip's design facilitates the rapid conversion of high complex AI models into real-time applications, enabling enterprises to harness the full potential of AI in their operations.
The Spiking Neural Processor T1 by Innatera is a revolutionary microcontroller designed to handle sensory processing with extreme efficiency. This processor is specifically crafted to operate at ultra-low power levels, below 1 milliwatt, yet it delivers exceptional performance in pattern recognition tasks right at the sensor edge. Utilizing a neuromorphic architecture, it processes sensor data in real time to identify patterns such as audio signals or movements, significantly outperforming traditional processing methods in both speed and power consumption. Engineered to function in always-on operation modes, this microcontroller is critical for applications where maintaining continuous operation is essential. Its design offloads processing tasks from the main application processor, allowing for dedicated computation of sensor data. This includes conditioning, filtering, and classification tasks, ensuring they are carried out efficiently within the strictest power limits. With its ability to be integrated with various sensors, the Spiking Neural Processor T1 empowers devices to achieve advanced functionalities such as presence detection, touch-free interfaces, and active monitoring in wearable devices. This product supports a comprehensive range of applications through its innovative approach to sensor data handling, leveraging the unique capabilities of spiking neural networks to drive cognitive processing in less power-intensive environments.
The HUMMINGBIRD by Lightelligence is an innovative optical Network-on-Chip processor that integrates photonic and electronic dies through advanced vertically stacked packaging technologies. This architecture provides a pathway to overcome conventional digital network limitations, particularly the 'memory wall.' With a 64-core domain-specific AI processor, HUMMINGBIRD uses a cutting-edge waveguide system to propagate light-speed signals, drastically reducing latency and power requirements compared to traditional electronic networks. This high-performance device serves as the communication backbone for data centers, facilitating data management and interconnect topology innovations. HUMMINGBIRD exploits the power of silicon photonics to offer a dense all-to-all data broadcast network that enhances the performance and scalability of AI workloads. HUMMINGBIRD's robust integration into PCIe form factors allows easy deployment onto industry-standard servers, and when paired with the Lightelligence Software Development Kit, it can significantly optimize AI and machine learning processes. This integration fosters a higher utilization of computing power and alleviates complexities associated with mapping workloads to hardware.
The Yitian 710 processor is a flagship Arm server chip developed by T-Head. It utilizes advanced architecture to deliver exceptional performance and bandwidth, supporting the latest Armv9 instruction set. Constructed with a 2.5D packaging, the processor integrates two dies, boasting a staggering 60 billion transistors. Designed for high-efficiency computing, it includes 128 high-performance Armv9 CPU cores. Each core encompasses a 64KB level one instruction cache, a 64KB level one data cache, and a shared 1MB level two cache. This architecture supports extensive on-chip memory including a 128MB system cache, ensuring rapid data access and processing.
aiWare represents a specialized hardware IP core designed for optimizing neural network performance in automotive AI applications. This neural processing unit (NPU) delivers exceptional efficiency for a spectrum of AI workloads, crucial for powering automated driving systems. Its design is focused on scalability and versatility, supporting applications ranging from L2 regulatory tasks to complex multi-sensor L3+ systems, ensuring flexibility to accommodate evolving technological needs. The aiWare hardware is integrated with advanced features like industry-leading data bandwidth management and deterministic processing, ensuring high efficiency across diverse workloads. This makes it a reliable choice for automotive sectors striving for ASIL-B certification in safety-critical environments. aiWare's architecture utilizes patented dataflows to maximize performance while minimizing power consumption, critical in automotive scenarios where resource efficiency is paramount. Additionally, aiWare is supported by an innovative SDK that simplifies the development process through offline performance estimation and extensive integration tools. These capabilities reduce the dependency on low-level programming for neural network execution, streamlining development cycles and enhancing the adaptability of AI applications in automotive domains.
The GenAI v1-Q from RaiderChip brings forth a specialized focus on quantized AI operations, reducing memory requirements significantly while maintaining impressive precision and speed. This innovative accelerator is engineered to execute large language models in real-time, utilizing advanced quantization techniques such as Q4_K and Q5_K, thereby enhancing AI inference efficiency especially in memory-constrained environments. By offering a 276% boost in processing speed alongside a 75% reduction in memory footprint, GenAI v1-Q empowers developers to integrate advanced AI capabilities into smaller, less powerful devices without sacrificing operational quality. This makes it particularly advantageous for applications demanding swift response times and low latency, including real-time translation, autonomous navigation, and responsive customer interactions. The GenAI v1-Q diverges from conventional AI solutions by functioning independently, free from external network or cloud auxiliaries. Its design harmonizes superior computational performance with scalability, allowing seamless adaptation across variegated hardware platforms including FPGAs and ASIC implementations. This flexibility is crucial for tailoring performance parameters like model scale, inference velocity, and power consumption to meet exacting user specifications effectively. RaiderChip's GenAI v1-Q addresses crucial AI industry needs with its ability to manage multiple transformer-based models and confidential data securely on-premises. This opens doors for its application in sensitive areas such as defense, healthcare, and financial services, where confidentiality and rapid processing are paramount. With GenAI v1-Q, RaiderChip underscores its commitment to advancing AI solutions that are both environmentally sustainable and economically viable.
The Universal DSP Library by Enclustra delivers streamlined solutions for digital signal processing within the AMD Vivado ML Design Suite. Providing a comprehensive library of DSP components like FIR and CIC filters, mixers, and CORDIC function approximations, the library enables developers to rapidly create signal processing chains. This is achieved through both raw VHDL source and Vivado IPI blocks, simplifying integration and significantly reducing development time. The library supports efficient bit-true software models in Python, allowing thorough evaluation of processing chains prior to FPGA implementation. This capability not only improves the development process but also provides a clear path from software simulation to hardware implementation. The DSP components are fully documented, with reference designs to guide users in combining various blocks to form complex DSP systems. Targeted at numerous applications such as software-defined radio and digital signal processing, the library addresses common DSP needs, freeing developers to concentrate on project-specific innovations. Furthermore, it supports multiple data channels and works with both continuous wave and pulse processing, utilizing the AXI4-Stream protocol for a standard interface structure.
The RISC-V Core IP from AheadComputing is engineered to deliver high performance while maintaining flexibility and efficiency in design. The open specification architecture allows users to tailor the core to meet diverse application demands, ensuring adaptability across various computing environments. This core IP is ideal for applications requiring customization and optimization, offering a robust solution for modern challenges in computing. Facilitated by a standards-based approach, AheadComputing’s RISC-V Core IP ensures seamless integration and compatibility, supporting a wide range of interfaces and functionalities. This extensibility makes it an excellent choice for projects where quick time-to-market and cost efficiency are critical factors. Moreover, the architecture is designed to support progressive enhancements and iterations, staying relevant in the fast-paced technology world. Particularly advantageous for embedded systems and consumer electronics, the RISC-V Core IP offers advanced processing capabilities without compromising on power efficiency. As the industry moves towards more open structures, this IP serves as a pivotal component in developing next-generation computing solutions.
The KL520 AI SoC by Kneron marked a significant breakthrough in edge AI technology, offering a well-rounded solution with notable power efficiency and performance. This chip can function as a host or as a supplementary co-processor to enable advanced AI features in diverse smart devices. It is highly compatible with a range of 3D sensor technologies and is perfectly suited for smart home innovations, facilitating long battery life and enhanced user control without reliance on external cloud services.
The AON1100 represents AONDevices' flagship in edge AI solutions aimed at voice and sensor applications. Its design philosophy centers on providing high accuracy combined with super low-power consumption. This chip shines when processing tasks such as voice commands, speaker identification, and sensor data integration. With a power rating of less than 260μW, the AON1100 maintains operational excellence even in environments with sub-0dB Signal-to-Noise Ratios. Its performance is highly appreciated in always-on devices, making it suitable for smart home applications, wearables, and automotive systems that demand real-time responsiveness and minimal energy draw. The AON1100 incorporates streamlined algorithms that enhance its sensor fusion capabilities, paving the way for smarter device contexts beyond traditional interactions. Its RISC-V support adds an additional layer of flexibility and compatibility with a wide range of applications, contributing significantly to the chip's adaptability and scalability across various domains.