Cadence Launches Tensilica Vision Q7 DSP: 2× AI & Vision Performance for Automotive, AR/VR, and Mobile

Cadence Design Systems has expanded its high‑end Tensilica Vision DSP family with the Cadence Tensilica Vision Q7 DSP, delivering up to 1.82 tera‑operations per second (TOPS). The sixth‑generation Vision Q7 is engineered to meet the escalating computational demands of embedded vision and AI workloads, offering up to twice the AI and floating‑point performance of its predecessor, the Vision Q6, without increasing silicon area.

Designed with simultaneous localization and mapping (SLAM) in mind, the Vision Q7 excels in robotics, drone, mobile, automotive, and AR/VR applications that rely on inside‑out tracking. Because SLAM requires both fixed‑ and floating‑point arithmetic, the DSP provides balanced, high‑throughput performance for both data types, enabling faster, lower‑latency edge SoCs.

For AI inference, the Vision Q7 supports 512 8‑bit multiply‑accumulate (MAC) units, double the 256‑MAC count of the Vision Q6. When paired with the Tensilica DNA 100 processor, the Q7 unlocks even higher AI throughput. The DSP also introduces advanced iDMA features—including 3‑D DMA, compression, and a 256‑bit AXI interface—boosting data movement efficiency.

The Vision Q7 is a superset of the Vision Q6, preserving existing software investments and simplifying migration from Vision Q6 or Vision P6 platforms. It supports AI frameworks such as Caffe, TensorFlow, and TensorFlow Lite via the Tensilica Xtensa Neural Network Compiler, which translates neural networks into highly optimized code for the Q7. The platform also implements the Android Neural Network API for on‑device AI acceleration and offers full, optimized support for over 1,700 OpenCV functions, accelerating migration of existing vision applications.

All development tools and libraries are built to help SoC vendors meet ISO 26262 ASIL D certification, underscoring Cadence’s commitment to safety and reliability in automotive and industrial markets.