Hexagon (processor)

Hexagon (processor)
Name	Hexagon
Designer	Qualcomm
Bits	32-bit, 64-bit
Introduced	2006
Design	DSP, VLIW
Application	Mobile devices, Internet of things

Hexagon (processor). The Hexagon is a digital signal processor (DSP) core designed by Qualcomm and integrated into its Snapdragon system on a chip (SoC) platforms. First introduced in 2006, it is engineered to handle specialized computational workloads like audio processing, computer vision, and artificial intelligence with high power efficiency. Its evolution has been central to enabling advanced features in modern smartphones and expanding into other connected devices.

Overview

The Hexagon DSP serves as a specialized co-processor within the broader Snapdragon architecture, offloading specific tasks from the main ARM CPU cores. This design philosophy improves overall system performance and energy efficiency for multimedia and sensor-based applications. It is a foundational component for technologies like computational photography, always-on voice assistants, and augmented reality experiences in devices from companies like Samsung and Google. The processor's role has expanded significantly with the rise of on-device AI inference and machine learning.

Architecture

The Hexagon architecture employs a very long instruction word (VLIW) design, allowing multiple operations to be issued in a single clock cycle for high throughput on DSP algorithms. Key architectural features have included the Hexagon Vector eXtensions (HVX) for SIMD operations critical to image processing and neural network acceleration. The core integrates tightly with other SoC components like the Adreno GPU and Spectra image signal processor through the Snapdragon heterogeneous compute framework. Over generations, the architecture has added support for hardware acceleration of tensor operations and improved scalar performance.

Development and history

Qualcomm introduced the first Hexagon core, known as the QDSP6, in 2006 within the MSM platform. A significant milestone was the 2013 introduction of the Hexagon 680 DSP in the Snapdragon 820, which featured the first generation of HVX for advanced imaging tasks. Subsequent generations, like the Hexagon 685 in the Snapdragon 835, began integrating a dedicated Tensor Accelerator for AI workloads. The architecture transitioned to a 64-bit design with later versions, and its development has been closely tracked at industry events like the annual Snapdragon Summit.

Applications and implementations

Hexagon DSPs are implemented across nearly all Qualcomm Snapdragon mobile platforms, from premium tiers in flagship devices like the Samsung Galaxy S series to entry-level chips for the Internet of things. They are instrumental in enabling features such as HDR photo capture, noise cancellation in VoIP calls, and real-time object detection for camera applications. Beyond smartphones, the technology is deployed in extended reality headsets, automotive infotainment systems, and robotics. The Snapdragon 8 Gen series showcases its most advanced implementations.

Performance and features

Performance is characterized by high computational density per watt, excelling at fixed-function and parallelizable tasks compared to general-purpose CPU cores. Key features across generations include advanced power management for always-on contexts, support for INT8 and INT16 precision for efficient AI inference, and hardware-accelerated computer vision libraries. Benchmarks often demonstrate orders-of-magnitude improvements in tasks like image recognition and natural language processing when leveraging the Hexagon DSP and its accelerators versus the ARM CPU alone.

Software and programming

Programming the Hexagon DSP typically involves using Qualcomm's Snapdragon Neural Processing Engine (SNPE) SDK and the Hexagon SDK, which provide tools for optimizing and deploying machine learning models. Developers utilize libraries like TensorFlow Lite and PyTorch Mobile that can delegate operations to the Hexagon backend. The Qualcomm AI Engine abstracts the heterogeneous hardware, including the Hexagon processor, for application frameworks. Support is also integrated into Android development environments for features like Android Neural Networks API.

Category:Qualcomm Category:Digital signal processors Category:ARM architecture Category:Mobile technology