Next generation Tensilica processor targets embedded AI

 Next generation Tensilica processor targets embedded AI

Technology News |
By Nick Flaherty

Cadence Design Systems has launched its next generation of extensible processor, aiming squarely at more power efficient embedded AI chip designs.

The Tensilica Xtensa LX8 processor platform is the foundation for the eighth generation of its industry-leading Xtensa LX processor family. This adds new digital signal processing (DSP) blocks, multi-processor, interconnect and system-level IP to meet the system-level performance and AI requirements of system on chip designs.

This comes as RISC-V is making significant in-roads into AI chip designs with extensible instructions, so Cadence has added its IP expertise as well.

The L2 cache enhancements see performance improvements of 50% or more for cache-based subsystems compared to the Xtensa LX7 processor while reducing pressure on L1 caches, while improved branch prediction achieves significant speedup in increasingly critical control-code performance.

An enhanced ARM AMBA interface supports both the original AMBA 4 AXI interconnect for easy integration into today’s high-performance devices and low-latency APB interface that can improve system performance while reducing pressure on the primary system bus

The enhanced iDMA IP improves 3D DMA transfers found in complex DSP algorithms while also adding compression/decompression support and expanding the physically addressable memory to 40 bits and the expanded interrupt support accommodates up to 128 interrupts to service the most demanding system-level requirements

Scaling processor performance to meet the ever-increasing performance demands of intelligent edge devices for mobile and automotive applications requires designers to look holistically at the system-level requirements. Factors such as data movement, memory bandwidth, latency and ease of integration all become more critical as SoC designs grow more complex.

For example, recurrent and convolutional neural networks often include large datasets that must be quickly accessed from system memory in order to meet real-time requirements. Processor and DSP subsystems must now support multiple concurrent algorithms and ever-larger width filters, so the ability to reduce the latency of memory accesses and limit DMA transfers of repetitive zeroes can greatly improve overall system performance.

These requirements, along with the continued emphasis on energy-efficient computing and AI, drove the development of the Xtensa LX8 platform, which includes several features designed to optimize overall system performance per watt.

The LX8 enhancements will be used across the broad family of Tensilica HiFi Vision, ConnX and FloatingPoint DSPs and AI controllers.

“We are pleased that Cadence is introducing a new generation of its Tensilica Xtensa LX product line,” said Jaeyoung (Jay) Jang, vice president of SoC at SK hynix memory solutions America.

“Having successfully utilized previous Xtensa LX processors in our SoC designs, we can confidently attest to their unparalleled efficiency and capability in application-specific processing. Cadence’s focus on system-level performance provides assurance that SoCs designed with Xtensa LX processors can meet end users’ demanding performance and power requirements.”

“Synaptics and Cadence have collaborated on numerous SoCs focused on AI-enabled edge devices. Leveraging the exceptional capabilities of Cadence Tensilica DSPs and Xtensa controllers, we have been able to create energy-efficient products catering to sensor, speech and vision applications, including cutting-edge biometrics, smart-home security and intelligent appliances,” stated Saleel Awsare, senior vice president and general manager, PC & Peripherals Division at Synaptics. “We’re enthusiastic about the evolution of Cadence’s new Xtensa LX8 platform, providing even more robust system-level performance to meet the demands of next-generation edge AI SoCs.”

“Today’s advanced SoC designs demand even greater processor subsystem performance. Processors based on the Xtensa LX platform are widely used today in the most demanding audio/voice, automotive ADAS and embedded computing applications, so we have extensive first-hand knowledge of our customers’ challenges in the drive for ever-increasing system-level performance,” said David Glasco, vice president of research and development for Tensilica IP at Cadence. “Key capabilities in the latest generation of our industry-leading extensible processor platform enable our customers to create even more advanced domain-specific processors.”

The Tensilica Xtensa LX8 processor is currently shipping to early access customers, with general availability expected in the late third quarter of 2023.

If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :    eeNews on Google News


Linked Articles