Target Compiler Technologies, the leader in EDA tools for the design and programming of ASIPs (application-specific instruction-set processors), announced several new improvements of its Chess/Checkers tool suite geared to the design of ultra-low power SoCs. Key to the innovation is multi-faceted support for parallelism as well as RTL-level optimizations common only in the most advanced design flows. The new tool capabilities will be demonstrated at the 44th Design Automation Conference in San Diego.
While there is an ever growing quest for more functionality and higher performance in today’s SoC designs, there is an even more pronounced need to minimize energy consumption – either to prolong battery life or to reduce operating temperatures. “Our customers tell us that following the operating frequency curve just isn’t a tangible solution any more. Advanced design teams realize that they must ‘get specialized’ and ‘go parallel’ in their design approaches to meet these seemingly conflicting requirements”, says Steve Cox, Target’s North American VP of Business Development.
The ASIP approach enables designers to specialize individual blocks of an SoC to their specific functional requirements – including introducing data-level and instruction-level parallelism to assure maximum computation per clock cycle. The MPSoC (multi-processor SoC) approach enables designers to target different functions to different ASIPs across the entire SoC, thereby introducing coarse-grained task-level parallelism and keeping new functionality from competing for processing cycles on the SoC’s primary embedded processor. Together, the ASIP and MPSoC approaches work to minimize power dissipation overall while also maximizing computational efficiency (measured as performance/$/watt) of a given SoC.
Overall, three important new functionalities have been added to the Chess/Checkers tool-suite:
First, new low-power RTL-level optimizations have been added to the hardware generation component of the Chess/Checkers tool-suite. The new hardware generator selectively inserts dedicated logic in the ASIP’s circuit to avoid unnecessary switching activity. The optimizer supports a user-controlled combination of clock gating, operand isolation, and optimized generation of register addresses. Measurements indicate that these new optimizations result in power savings of more than 60% compared to the previous Chess/Checkers release, and in power dissipation metrics that are within a few percentage points of equivalent RTL designs that were hand-optimized by low-power hardware specialists. With these optimizations, Chess/Checkers now produces the quality of results needed to effectively bridge the gap from architectural exploration to RTL implementation.
Secondly, the instruction-set simulator of the Chess/Checkers tool-suite has been extended with new fast instruction-accurate simulation techniques. The new simulation technology delivers a measured simulation speed of many tens of MIPS for complex DSP architectures – over 100 times faster than conventional cycle-accurate simulators. Such simulation speed is a key enabler for MPSoC designs, enabling speeds suitable for virtual prototyping, even when multiple processors are included in a single co-simulation.
Finally, enhanced support for instruction predication has been added to the optimizing C compiler component of the Chess/Checkers tool-suite. With instruction predication, new and more powerful parallelism in ASIPs is possible – unleashing greater degrees of both instruction-level parallelism (for VLIW architectures) and data-level parallelism (for SIMD architectures). SIMD architectures are common in video and image processing, as well as wireless modem applications, all of which are particularly sensitive to power concerns in this age of mobile personal devices.
“The era of the ‘power-envelope’ is here. Whether targeting battery-powered devices or devices with modest thermal requirements, today’s designers need tools that can meet performance requirements within a pre-defined power budget.” said Gert Goossens, CEO of Target. “These new capabilities add to our already well known capabilities in ASIP architectural exploration and optimization – thus extending Target’s lead in ultra-low power, high performance SoC design.”
The new RTL optimizations and support for instruction predication are available in the current release of Chess/Checkers. The fast simulation capability is in beta test at customer sites now and will be released later this year. All of these new features will be made available to existing customers as a maintenance upgrade.
About Target Compiler Technologies
Target Compiler Technologies is the leading provider of retargetable software tools to accelerate the design, programming and verification of application-specific processor cores (ASIPs). Target’s Chess/Checkers tool suite has been applied by customers worldwide for diverse application domains, including GSM, WCDMA and HSDPA handsets, VoIP, audio coding, car infotainment, ADSL and VDSL modems, wireless LAN, hearing instruments, mobile image processing, video processing, and various control and interfacing applications. Target is a spin-off of IMEC, is headquartered in Leuven, Belgium, with North American operations in Boulder, Colorado.