feature article
Subscribe Now

Choices On Top of Choices

Tensilica Fusion DSP Core Geared for Low-Power Devices

“Embedded” has a new spelling, and it’s written IoT.

Tensilica is a company that made its name as an early pioneer of tweak-it-yourself microprocessors. Actually, the name itself is a play on the words “tensile” (as in, stretchable) and “silicon,” which pretty well encapsulates the company’s unique selling proposition. Its microprocessor IP is just like any you’d get from ARM or MIPS or anyone else – except that it’s not. Instead of delivering you a prepackaged CPU core, Tensilica instead hands you a configuration tool. Pick your instruction set, pick your bus width, pick your execution resources, and the Tensilica tool will generate a custom CPU for you on the fly. It’s the whiteboard of processor product roadmaps.

EDA giant Cadence liked Tensilica so much it bought the company. Which kinda makes sense, since most EDA companies also supply IP, and Tensilica was already kind of an EDA tool company anyway. Synopsys acquired fellow configurable-processor company ARC, and Mentor Graphics acquired Accelerated Technology, Microtec Research, and others. It’s what big EDA companies do.

So now that Tensilica is a brand underneath the Cadence umbrella, it’s started to roll out new CPU core designs they have created post-acquisition. And the first of these is Fusion. Fusion is really more of a DSP than a conventional RISC CPU, and it’s intended – like everything these days – for the dreaded Internet of Things. (How long before our descendants make fun of us for using that phrase? “Gee, Gramps, did you also have the ‘Highway of Cars’ or the ‘River of Boats?’”)

Because Fusion is IoT-bound, it’s designed for efficiency, small die area, and low power. None of these characteristics are quantifiable, of course, because we’re talking about IP, not finished chips, and user-configurable IP at that. So objective measurements are mostly nonexistent.

What Fusion does do is to pare down some of the more exotic configurability options that Tensilica’s other CPU cores support, and those that wearable IoT gadgets presumably don’t need. But wait – if it’s user-configurable, couldn’t you get the same result by making those same changes to an existing Tensilica CPU? Well, yes and no. Tensilica’s configurability has limits, and the same CPU framework doesn’t work ideally in all situations. Thus, Tensilica offers separate varieties of its CPU prewired for different application areas, and then lets you tweak it from there. It’s a bit like selecting the truck, the sports car, or the sedan and then haggling over the options. Narrow down the choice first, and then fine-tune the preferences. Fusion is the choice for data-acquisition and/or wireless communication on a budget.

Being DSP-like, Fusion has a quad MAC, which can be configured (naturally) a number of different ways, from dual 24×24 to single 32×32, with variations in between. Fusion also sports a single-precision floating-point unit, with a dual-issue pipeline that can dispatch integer operations at the same time as FP ops.

Fusion is also equipped to handle crypto work, including AES-128, but it’s not done by a dedicated hardware cryptography block. Instead, the CPU processes crypto operations as part of its basic instruction set, should you choose to configure it to do so. That’s part of the fun of a user-configurable processor: you get to decide what it’s good at. Baseband bit operations are handled the same way.

The real magic of a user-configurable processor isn’t really in the hardware; it’s creating a compiler and other software tools that don’t break every time the customer changes something. Tensilica has this all figured out, of course. The compiler is auto-generated at the same time that the hardware netlist is, so they always match. And there are certain core features and instructions that you can’t remove. In other words, the configuration tool prevents you from creating a nonworking processor, or one that’s incompatible with everything else in the world. A stripped-down, bare-minimum configuration still runs real code and real operating systems. You get to add options on top of that, including your own home-grown features and instructions, if you wish. After making your design choices, you push the big red GO button, and Cadence’s big computer back at headquarters cranks out your custom processor, along with the relevant software tools. 

So now your IoT processor can be different from everyone else’s IoT processor. Even if the marketing message is probably just the same. 

Leave a Reply

featured blogs
Jul 25, 2021
https://youtu.be/cwT7KL4iShY Made on "a tropical beach" Monday: Aerospace and Defense Systems Day...and DAU Tuesday: 75 Years of the Microprocessor Wednesday: CadenceLIVE Cloud Panel... [[ Click on the title to access the full blog on the Cadence Community site. ]]...
Jul 24, 2021
Many modern humans have 2% Neanderthal DNA in our genomes. The combination of these DNA snippets is like having the ghost of a Neanderthal in our midst....
Jul 23, 2021
Synopsys co-CEO Aart de Geus explains how AI has become an important chip design tool as semiconductor companies continue to innovate in the SysMoore Era. The post Entering the SysMoore Era: Synopsys Co-CEO Aart de Geus on the Need for AI-Designed Chips appeared first on Fro...
Jul 9, 2021
Do you have questions about using the Linux OS with FPGAs? Intel is holding another 'Ask an Expert' session and the topic is 'Using Linux with Intel® SoC FPGAs.' Come and ask our experts about the various Linux OS options available to use with the integrated Arm Cortex proc...

featured video

Design Success with Foundation IP & Fusion Compiler

Sponsored by Synopsys

When is 1+1 greater than 2? When using DesignWare Foundation IP & Fusion Compiler! Join Raymond and Yung in their discussion of a customer that benefited from the combination of Fusion Compiler’s machine learning and Foundation IP cells and macros.

More information about DesignWare Foundation IP: Embedded Memories, Logic Libraries, GPIO & PVT Sensors

featured paper

Long-term consistent performance matters for humidity sensing applications

Sponsored by Texas Instruments

The exposed polymer of humidity sensors can be impacted by the environment, leading to drift over time. This article from Texas Instruments discusses the accuracy and long-term drift of humidity sensors and how these parameters affect system performance and lifetime.

Click to read more

featured chalk talk

SN1000 SmartNIC

Sponsored by Xilinx

Cloud providers face a variety of challenges with moving data from one place to another. In modern data centers, flexibility is a key consideration - on par with performance. Software-defined hardware acceleration offers a major breakthrough in flexibility. In this episode of Chalk Talk, Amelia Dalton chats with Kartik Srinivasan of Xilinx about the details of Smart NICs with the new Alveo SN1000 with composable hardware.

Click here for more information about the Alveo SN1000 - The Composable SmartNIC