XCELLENCE IN PROTOTYPING Lowering the Barriers <strong>to</strong> a Successful FPGA-Based Pro<strong>to</strong>type Project The <strong>Xilinx</strong> Virtex-6 and Synopsys HAPS systems make pro<strong>to</strong>typing practical for ASIC designs. by Troy Scott Senior Product <strong>Market</strong>ing Manager Synopsys troy.scott@synopsys.com 32 <strong>Xcell</strong> <strong>Journal</strong> First Quarter 2012
Recent surveys of the ASIC designer community show that more than 70 percent of all design projects use some form of FPGA-based pro<strong>to</strong>type <strong>to</strong> provide a high-speed model of the design prior <strong>to</strong> the tapeout of a test chip. The pro<strong>to</strong>type helps <strong>to</strong> start software development tasks and <strong>to</strong> close functional coverage earlier in the design cycle. In many organizations, the versatility and reuse benefits of a commercially available FPGA-based pro<strong>to</strong>typing system have led <strong>to</strong> a much improved return on investment compared <strong>with</strong> cus<strong>to</strong>m circuit boards developed in-house. The Synopsys HAPS High-Performance ASIC Pro<strong>to</strong>typing System has a long pedigree of using <strong>Xilinx</strong> ® Virtex ® FPGAs as a logic host for ASIC pro<strong>to</strong>typing. Starting in 1987 at HARDI Electronics AB of Lund, Sweden (now part of the Hardware Platform Group at Synopsys), the HAPS system of circuit boards has proven <strong>to</strong> be a versatile <strong>to</strong>ol for thousands of pro<strong>to</strong>typing projects. To understand why, it’s instructive <strong>to</strong> take a deeper look at the functional elements of the Virtex-6 and the HAPS system architecture that have made this combination a state-of-the-art solution. VIRTEX-6 IS IDEAL FOR ASIC PROTOTYPING Throughout its his<strong>to</strong>ry the HAPS pro<strong>to</strong>typing systems have leveraged the Virtex family of devices, including the Virtex-II, Virtex-II Pro, Virtex-4, Virtex- 5 and, most recently, Virtex-6 LX760 devices. The highest-capacity LX760 has proven its utility as a vehicle for ASIC pro<strong>to</strong>typing for many companies. Designers can apply both the programmable and embedded-block functions of the device <strong>to</strong> the pro<strong>to</strong>typing effort. The primary LX760 logic building block, a slice, is composed of programmable lookup tables (LUTs) and s<strong>to</strong>rage elements used <strong>to</strong> implement combina<strong>to</strong>rial functions, along <strong>with</strong> small RAM blocks or shift registers. A <strong>to</strong>tal of 118,560 slices are available in the LX760, providing a deep resource <strong>to</strong> host logic functions such as parity, XOR, AND, OR and synchronous logic of the original ASIC RTL. When considering the migration of arithmetic functions such as multipliers, accumula<strong>to</strong>rs and other DSP logic, the LX760 provides embedded signal-processing blocks that are far more area-efficient than slice-based logic cells. An LX760 provides 864 DSP blocks along <strong>with</strong> the DSP48E1 slice, consisting of a 25 x 18-bit two’s complement multiplier and 48-bit accumula<strong>to</strong>r. The Synopsys Synplify Premier FPGA logic-synthesis <strong>to</strong>ol au<strong>to</strong>matically targets these essential building blocks and provides the best clock performance. System-on-chip (SoC) designs include various memories provided by either the target ASIC fab’s memory library or a memory compiler. In most cases FPGAs can represent these memories efficiently using slices for smaller register files, embedded block RAM for wider, deeper arrays or, for the largest memories, HAPS memory daughterboards as a host. Effective ASIC memory migration is part of the FPGA-based pro<strong>to</strong>typing methodology, which often involves assigning an alternative FPGA-friendly implementation <strong>to</strong> ensure the most efficient use of onchip RAM. Substitutions—driven, for example, by Verilog HDL ‘define—help provide an easy way <strong>to</strong> switch between modules as the code is targeted <strong>to</strong> different implementations. In many cases ASIC memory compilers can generate an “emulation-friendly” light version of the model that excludes test logic so that the pro<strong>to</strong>typing can more easily migrate <strong>to</strong> the pro<strong>to</strong>type system. A Virtex-6 760LX provides up <strong>to</strong> 8,280 kbits of distributed slice-based RAM and 25,920 kbits of block RAM for hosting memory modules. XCELLENCE IN PROTOTYPING The Virtex-6 family provides up <strong>to</strong> nine embedded PLL-based mixed-mode clock managers (MMCMs) that can serve as frequency synthesizers for a wide range of frequencies, as a jitter filter for either external or internal clocks, and as deskew clocks. The Virtex-6 MCMM is capable of output frequencies from 4.69 MHz <strong>to</strong> 800 MHz. This is a key resource for the pro<strong>to</strong>type Designers can apply the programmable and embedded-block functions of the LX760 <strong>to</strong> the ASIC pro<strong>to</strong>typing effort. <strong>to</strong> achieve multimegahertz speeds of internal clocks. The Synopsys Certify multi-FPGA pro<strong>to</strong>typing environment eases the implementation of ASIC clocks by targeting both onboard PLLs of the HAPS system and the embedded LX760 MMCMs <strong>with</strong> clock-distribution IP that’s generated based on how a design is partitioned. HAPS’ MODULAR ARCHITECTURE APPLIED The HAPS design team applied some key design criteria <strong>to</strong> balance the cost, performance and connectivity needs of a pro<strong>to</strong>typing circuit board. For hardware-assisted verification <strong>to</strong>ols such as an ASIC emula<strong>to</strong>r, RTL debugging and verification was the primary application. But the pro<strong>to</strong>type community was more concerned <strong>with</strong> “validation,” a process in which testing scenarios are vastly expanded. This required that the pro<strong>to</strong>type performance be high enough <strong>to</strong> interface <strong>to</strong> real-world pro<strong>to</strong>cols like Gigabit Ethernet and double-datarate (DDR) memory interfaces, and execute SoC processors <strong>to</strong> boot an operating system. These implications convinced the hardware designers <strong>to</strong> provide a very open, modular architecture that maximized FPGA I/O access, along <strong>with</strong> a reconfigurable interconnect scheme First Quarter 2012 <strong>Xcell</strong> <strong>Journal</strong> 33