feature article
Subscribe Now

Accelerating Processor-based Systems

Designing an efficient processor-based system architecture with overall system performance optimized for a specific application is not trivial, requiring skills and technology similar to those employed by supercomputer designers. Accomplishing this feat requires new tools and methodologies to augment the EDA flow for both traditional ASIC design and the new class of programmable SoCs, e.g., FPGAs. Architectures and performance must be verified early in the design cycle; the designer cannot wait until RTL development to discover their architecture does not support their system requirements.

Problems to Solve

Current processor tools have not kept up with the challenges of the new high-performance systems. Designers need tools to explore and exploit performance, power, and cost opportunities in their processor-based designs. Designers need tools to help them address and overcome:

–Inadequate processor performance to support an ever-increasing functional demand;

–Memory hierarchy and system architecture bottlenecks;

–Increasing design scalability, i.e., enhancing the capability of the design without radical redesign work.

Tools that solve these problems will enable designers to quickly model the system architecture and explore tradeoffs between performance, power consumption, and cost.

Inadequate Processor Performance

To gain more headroom for an application or add extra features to an existing design, the designer has to find ways to get the code to run faster on the system. Visibility into the system to determine where the cycles are going and where inefficiencies exist is critically important to accomplishing this. Unfortunately, compilers and simple profilers cannot comprehend the more complex components and system architectures found in modern designs.

Many components today utilize special modes and operating conditions to increase their performance by optimally handling data and computation. These work well as long as the proper conditions are met, however large penalties are incurred when operations deviate from these conditions. An example of this is DDR memories, which provide very high data bandwidth. But the designer must properly account for the conditions and operating modes that DDR memories require to provide this bandwidth. Tailoring the code so it uses these resources best can often double the speed of code over the best optimizing compiler.

Software developers need a design environment that models these components so they can use the hardware more efficiently. By accurately simulating how the system will execute the code, the designer can then make the appropriate changes in data layout, array sizing, function partitioning, etc. The resultant system will not only utilize the available resources more efficiently, but will execute faster.

This is also critically important for derivative designs that add features and capabilities to existing code that must run on an existing platform. In most cases, the developer does not have the luxury of increasing the clock rate or moving to a newer, faster architecture. Therefore it is important to optimize the existing code and eliminate the inefficiencies not only inside the software, but between the software and the hardware as well.

Evaluating System and Memory Architecture

The rising complexity of SoCs makes system-level design a critical function shared by a large portion of the design team. At the same time, programmable platforms such as structured ASICs and FPGAs that allow the system designer to modify the architecture to best implement the code are becoming pervasive. Given the flexibility of the processor architecture and the prevalent use of legacy blocks and third-party IP, it is important to develop a sound system architecture that interfaces all of the externally and internally generated IP and satisfies all of the software requirements.

Modern processor based-design also offers an abundant selection of caches, memory sizes, bus topologies, etc., from which to choose. Working one’s way through these options and understanding the tradeoffs in the context of one’s application can be a daunting task without a tool to help objectively measure and compare them.

Furthermore, although various IP now come ready to connect to popular busses to simplify assembling them, the designer must properly set up their bus capacity and topology to weave them into an efficiently functioning system. How well a design team handles these challenges may well determine the success of the entire design program.

Design Scalability

As a product evolves, performance, power, and/or cost pressures often require the designer to scale the original design in one of these dimensions. Frequently, the solution is to move software functions into hardware, so as to:

–Increase performance – A dedicated hardware block will greatly accelerate the software function;

–Lower power consumption – A hardware block can use a much lower clock rate than an equivalent software-implemented function to get the same level of performance, thereby reducing power consumption;

–Lower system cost – A hardware block can often eliminate the need for one or more processors while still meeting product performance requirements;

–Protect IP – A hardware block is much more difficult to reverse engineer than software, which is subject to straightforward code disassembly and reverse engineering.

Traditional methods for moving functions to hardware required major effort on the part of the system designer. These require architecting new data flows through the system and developing a new software flow with an interface to the hardware. These programs require an expertise in processor design as well as the time to develop and test the additional hardware. Designers need a tool to provide scalability in their designs. This tool must provide a quick and easy flow to evaluate the system design and then generate additional hardware to support the new requirements.

Poseidon’s Triton Tool suite enables the designer to quickly develop an optimal architecture. The tool suite was created specifically for processor-based systems that require efficient robust architectures with the need to optimize performance, power and cost. Triton consists of two main tools:

–Triton Tuner – a system and software analysis tool;

–Triton Builder – a hardware accelerator synthesis tool.

Triton Tuner is a simulation and analysis environment based on SystemC. Tuner co-simulates the hardware and software system, identifying inefficiencies in the design. Triton Builder synthesizes algorithm-aware accelerator blocks to allow the designer to move algorithms from software to hardware. Together, these tools provide a system design and optimization environment that unobtrusively plugs into an existing design flow, enabling designers to quickly make dramatic improvements in the performance and power consumption of their application.

Design Flow

The Triton Tools allow the designer to use either tool independently, or both together as an integrated suite. A typical system design flow is usually an iterative process where the user analyzes the system performance, determines inefficiencies, modifies the system then checks the resultant performance. The Triton tools aids in this process by providing tools that accelerate the process of identifying the problem areas, and an integrated flow that allows the user to move between the tools to develop the optimal architecture. The typical flow (Figure 1) consists of:

–Triton Tuner profiles the ANSI C code, reveals bottlenecks in the code or architecture and eliminates the inefficiencies;

–Triton Builder partitions processor-intensive algorithms in ANSI C into hardware and generates a hardware accelerator;

–Triton Tuner verifies that the new system performs to the desired level.

Figure 1 – Poseidon Tools Flow

Tuner takes in the application source code in ANSI C and an architectural description. The user then performs system analysis and optimization on the simulated system. If additional performance gains are required, the designer uses Builder to further analyze and partition the system. The architecture and source code are then transferred into Builder, which generates a hardware accelerator moving selected loops and functions into hardware. Builder generates: modified source, transaction-level models of new hardware, drivers, test vectors and RTL for the accelerator. The designer then uses Tuner to verify the new design meets the system requirements.

Interface to Existing Tools

Triton Tools augment existing design tools through standard interfaces, including:

–ANSI C – Tuner and Builder accept and generate ANSI C for software descriptions.

–Compilers – Tuner and Builder use the existing compilers and build environment; Triton does not force the designer to adopt a new compiler. All Builder-created drivers follow standard ANSI C and pass through the existing flow.

–Synthesizable VHDL and Verilog RTL – All Builder-generated RTL is synthesizable VHDL or Verilog RTL that common synthesis tools such as Synplicity will accept.

–VHDL and Verilog Testbench – The Builder-generated testbench is compatible with standard VHDL and Verilog simulators.

Driver Generation and Integration

Builder simplifies the task of modifying the code to utilize the new hardware accelerator. The tool generates all of the drivers necessary to utilize the new hardware. The application code is also modified to reflect the new partition. Builder takes the original source code and creates a new copy of the application code with the algorithms that were moved to hardware extracted and the drivers inserted to invoke and control the accelerator. The tool provides the designer with executable code that runs on the new hardware and matches the functionality of the old architecture. The generated drivers contain:

    –Downloads of the code for the accelerator control unit. This is commented machine code so the designer can update it if they wish.

    –Storage of the accelerator parameters for the particular execution of the accelerator, including pointers to data location, coefficient updates, etc;

    –Trigger for the accelerator execution;

    –Polling loop, if the designer selects polling as the method to monitor progress;

    –Loading from the accelerator parameters and status from execution.

The driver makes use of the accelerator transparent to the rest of the software. The designer can use either polling or interrupts for notification of accelerator completion.

Testbench and Verification

Builder creates the testbench to verify the functionality of the accelerator. The design can then be evaluated and verified in Tuner. Builder creates the necessary files and models to execute the updated system with the accelerators using the high-speed transaction-level simulation in Tuner. This enables the designer to verify the performance of the new architecture and verify the functionality.


In short, Poseidon’s Triton Tool suite enables the designer to add sophisticated hardware acceleration to their processor-based system. This enables designers to achieve higher system throughput, reduced power consumption and cost as well as shorten design cycle.

Leave a Reply

featured blogs
Aug 18, 2018
Once upon a time, the Santa Clara Valley was called the Valley of Heart'€™s Delight; the main industry was growing prunes; and there were orchards filled with apricot and cherry trees all over the place. Then in 1955, a future Nobel Prize winner named William Shockley moved...
Aug 17, 2018
Samtec’s growing portfolio of high-performance Silicon-to-Silicon'„¢ Applications Solutions answer the design challenges of routing 56 Gbps signals through a system. However, finding the ideal solution in a single-click probably is an obstacle. Samtec last updated the...
Aug 17, 2018
If you read my post Who Put the Silicon in Silicon Valley? then you know my conclusion: Let's go with Shockley. He invented the transistor, came here, hired a bunch of young PhDs, and sent them out (by accident, not design) to create the companies, that created the compa...
Aug 16, 2018
All of the little details were squared up when the check-plots came out for "final" review. Those same preliminary files were shared with the fab and assembly units and, of course, the vendors have c...
Jul 30, 2018
As discussed in part 1 of this blog post, each instance of an Achronix Speedcore eFPGA in your ASIC or SoC design must be configured after the system powers up because Speedcore eFPGAs employ nonvolatile SRAM technology to store its configuration bits. The time required to pr...