The New Tensilica ConnX 110 and 120

This morning, we announced the two latest members of the Tensilica ConnX family, targeted at the automotive, consumer, and industrial markets. The two new members of the family are the ConnX 110 and ConnX 120 DSPs. They share a common instruction set architecture with proven high-performance ConnX B10 and B20 DSPs, but are lower power, smaller, and obviously, less powerful (in the computational sense, not just the watts sense).

For my previous coverage of the ConnX DSP family, see my posts:

• Tensilica ConnX B20 for 5G, and Automotive Radar/Lidar when we announced the B20 at a press conference at embedded world in 2019, which I attended and was the last time the event was held in person until earlier this month, which I also attended. See my post Embedded World 2022.
• Tensilica ConnX B10 in GF 22FDX for Automotive Grade 1, where GlobalFoundries used the B10 as a test vehicle.
• Implementing Automotive Radar on Tensilica Processors about implementing frequency modulated continuous wave (FMCW) radar with the ConnX BBE32EP.
• Cadence/Samsung Automotive Reference Flow which used the B10 as the test vehicle for a certified automotive reference flow.

The 128-bit ConnX 110 DSP and 256-bit ConnX 120 DSP feature an N-way programming model and are fully compatible with the ConnX B10 and B20 DSPs, preserving software compatibility for easy migration. They are also fully-paid-up members of the Tensilica product family, so implement the Tensilica Instruction Extension (TIE). For more on that, see my post Custom Instructions in Tensilica: Wearing a TIE Makes You Smarter.

The new DSPs are supported by a comprehensive set of complex math library functions in the NatureDSP, Eigen, and Radar libraries. All ConnX DSPs are automotive-ready with full ISO 26262 compliance to ASIL-D with FlexLock or to ASIL-B.

The Tensilica ConnX 110 and 120 DSPs offer:

• Optimized instruction set for radar, lidar, and communications applications
• 128-bit (ConnX 110) and 256-bit (ConnX 120) SIMD performance for complex math operations based on 8-, 16- and 32-bit fixed-point and half-, standard- and double-precision floating-point
• Common instruction set and simplified programming model, allowing code to be written once and utilized across different SIMD widths within the ConnX family
• Architecture optimized for small memory footprint and low-power signal processing
• Optional acceleration operations for linear-feedback shift, convolutional encoding, single peak search, and dual peak search
• The ConnX 120 additionally offers Viterbi and Turbo decoders

One of the companies quoted in the press release is NXP, which you probably know used to be Philips Semiconductors, which acquired my long-time previous employer VLSI Technology in 1999 before being spun out under the NXP name in 2006. It acquired Freescale in 2015. It is one of the biggest companies in automotive semiconductors and has made extensive use of the prior Tensilica models, although not the ConnX 110 and 120 since they are brand new. Robert Dunnigan is director of program management for ADAS there and tells the story:

NXP continues to rely on the Cadence Tensilica DSP cores for our ADAS product offerings. We have several generations of radar products covering applications from simple corner radar to advanced front end and 4D imaging radar solutions that all utilize these DSP cores. We are looking forward to the advancements in feature set and performance improvements that the ConnX family of products will bring, as well as the continued focus on Automotive ASIL-B/D support and ISO 26262 compliance that Cadence provides with their DSP offerings.

Sign up for Sunday Brunch, the weekly Breakfast Bytes email.

.