Allan Cantle knows what it takes to design high-performance digital systems. He’s been proving it since his early days at BAE Systems, where he was creating real-time simulations of evasive enemy targets to be used with real-world missiles in targeting practice. Allan has always had a knack for decomposing a complex problem into manageable-sized chunks, then mapping those chunks onto the appropriate architectures with the right interconnect to hit aggressive performance goals in the most economical way. Allan has a passion for working on complicated computing problems and isn’t afraid of trying varied technologies to solve them. When FPGAs came into his world, he knew just what to do with them. He understood that the highest computational throughput always comes from molding the architecture to the algorithm, and not the other way around.
Allan founded Nallatech (www.nallatech.com) in 1993, but he’d been busy building his alternative concept of supercomputing long before that. Allan comes from a family with a long history of driving technological innovation. His grandfather was an electrical pioneer as well. When electricity first became available in his Bristol, England hometown, the electric company required at least 20 houses to sign up in an area before they’d bring in service. Wanting electricity for his own home, he canvassed the area and wired enough neighboring houses himself to meet the minimum requirement. The same inventive and entrepreneurial spirit carried on to Allan’s father, and to him.
For Allan’s 14th birthday, his father, who was an avid tinkerer and repairman, gave Allan a Sinclair Computer-in-a-Bag (figure 1) as a learning project. Allan (figure 2) assembled it and became hooked on the idea of computers and building electronic hardware. Allan finished school in Britain at age 16 at the “O” (ordinary) level rather than the “A” (advanced) level normally required for university entrance. He went directly from school to work, beginning a four-year apprenticeship at what is now BAE Systems. “The apprenticeship rotated me through various departments, including engineering, accounting, and several others,” Allan recalls. “The first year was almost purely education, and the second year I landed in the computing resources department. All the variety was actually very good preparation for running a business. I got first-hand experience in a number of different areas in the company.”
Figure 1. Sinclair Computer-in-a-Bag, c. 1980
Figure 2. Allan Cantle, c. 1980
The area where Allan thrived the most, however, was computing. “At the time, our computers were IBM mainframes and DEC mini-computers,” Allan continues. “Initially, these computers were ‘alien’ to the engineers and physicists who had complicated problems to solve. Their primary tools were paper, pencils, and calculators, so we had a team of about fifty programmers who would partner with them. The programmers would help the engineers and physicists use the computers on their projects. I quickly felt that I belonged and could perform in this work.”
Over time, however, programming became part of the new engineers’ competence, so the need for the dedicated programming group went away. Allan moved into a more hardware-intensive role, developing printed circuit boards that plugged into DEC Q-bus backplanes for measurements on video signals used in weapons systems. “These projects ended up being very specialized high-performance computing systems,” Allan explains. “We developed several versions of a target simulation for the Rapier air-defense missile system.” The system required massive computing power, as it was inserted into the video feed from a real missile and simulated a realistic target. Its job was to create a realistic simulated image of an incoming hostile target, complete with smoke plume, so that the team could evaluate the performance of actual missiles in tracking and hitting targets without having to use a real target drone. “At first, we had to buffer the images of the simulated target, but over time we got enough computing performance to generate the images in real-time, including simulated effects of weather and degradation from the optics of the actual camera.”
Allan’s extraordinary success with the project got the attention of management, who strongly suggested that he get himself to University to get a degree. Allan finished his degree in 1988 and founded Nallatech as a part-time business in 1993. “Water service was being privatized, and I designed a system to control a water distribution center,” Allan continues. “I designed a beautiful prototype system that worked perfectly at the customer’s test site. They then took it and installed it in the field. They called me soon after and asked for help, telling me that the controller was crashing a lot. When I arrived, I found my controller mounted in between two giant, high-voltage electrical main contactors that generated a great deal of RF noise. I was amazed that it had worked at all. I went back and worked on a shielded/hardened version, but by the time we were ready, the opportunity had passed, and I decided to lay Nallatech dormant.”
Allan left BAE to relocate to Scotland in 1995, following his wife’s work, and re-established Nallatech there as a full-time engineering consultancy. “After using FPGAs extensively as co-processors for computing in BAE, I decided to focus Nallatech’s business around FPGA technology and established a partnership with industry leader Xilinx. That relationship has contributed to our success over the last ten years,” Allan explains. “With Xilinx’s support and tools, we were able to land several big contracts, including some with my former employer, BAE Systems. We have developed several generations of their XtremeDSP Development Kit that the industry has embraced. Our success pulled in some big orders for Xilinx as well, so it turned out to be a win-win relationship.”
In 1996, Allan went to an entrepreneurship program in Scotland, where he first met Dr. Malachy Devlin. Allan and Dr. Devlin hit it off immediately, and Allan tapped Malachy to come to work as Nallatech’s CTO. “We sat next to each other at the program purely by coincidence,” Allan recalls. “He had a software-centric background and understood hardware. I had a hardware-centric background and understood software. He was an innovative, creative, free-thinking type, and I was a pragmatic, execution-focused engineer. We balanced each other perfectly.”
Leveraging their combined expertise, the two steered Nallatech down the path of demonstrating the incredible computing power that could be delivered by FPGAs in expert hands. They amassed a string of impressive successes in video and image processing and other high-performance, compute-intensive problems. Their powerful results soon put their services in high demand, and they looked for ways to scale their operation in order to serve a broader customer base. “In 1997-1998 we started to notice that a lot of our designs had major parts in common,” Allan observes. This observation led to the development of the company’s original DIME FPGA Computing architecture and latterly to enabling a set of supporting tools that includes DIMEtalk. DIMEtalk is an application-development environment for FPGA computing hardware, with support for a variety of leading high-level FPGA design tools and an integrated C-to-VHDL compiler. It abstracts the features of the hardware platform from the development environment to accelerate one of the most time-consuming elements of FPGA system design. At the simplest level, DIMEtalk is like virtual, easy-to-use Lego blocks for Nallatech’s FPGA-based design modules and systems.
Over the following years, Nallatech has developed a range of hardware modules and software support systems to simplify the rapid design and deployment of ultra-high performance computing applications with FPGAs. The efficiency and experience they gained, combined with consistent, creative, engineering success, resulted in explosive growth for the company. They went through two rounds of funding, led by 3i and Scottish Equity Partners, SEP, to fuel the expansion they needed to handle the new business their success was breeding. The fledgling startup had grown into the world’s technology leader in specialized embedded FPGA-based computing systems.
Now at over 60 employees and growing, Nallatech is turning another corner. This year, they’ve announced a series of new partnerships aimed at extending their reach in Embedded COTS solutions and establishing “street cred” in the elusive high-performance computing (HPC) business. This represents a nice piece of forward thinking – the kind that has generated most of Nallatech’s success thus far. Like traditional HPC leaders SGI, Cray, and SRC, Nallatech believes that FPGAs form the future of high-performance computing. Compared with many of those players, however, Allan Cantle has an alternative view of the ideal HPC architecture. “We believe an FPGA-centric system with stand-alone processing will eventually dominate high-performance computing,” Cantle predicts.
Traditional HPC companies have been grafting FPGA-based accelerator modules onto their stalwart parallel processor architectures. The FPGAs, they reason, allow offloading of massively parallel tasks into dedicated hardware while the traditional processors keep up with the control-dominated load. While this sounds great in theory, it’s very difficult to get it to execute anywhere near its potential in practice, and the difficulty of programming the hybrid hardware is a deterrent to all but the most determined customers. While the FPGAs can create datapaths with enormous throughput, the designs typically fall short in the communication path between the FPGAs and the core processors, or in the decisions behind the hardware/software partitioning, or in the quality of hardware generated by automatic algorithmic synthesis tools.
This is where Allan Cantle’s Nallatech enters the picture. Nallatech’s experience designing dedicated HPC hardware for specific algorithms made them an ideal collaborative partner for Silicon Graphics Inc. (SGI) when they went looking for help in the high-end FPGA space. Their collaboration, announced this July, involves joint development of Reconfigurable Application-Specific Computing (RASC) solutions for high-value applications requiring enormous computational capability. The combination of SGI’s high-bandwidth shared-memory HPC architecture, market position, and distribution channel with Nallatech’s cutting-edge FPGA-based computing expertise will likely create a formidable competitor in their targeted markets.
The question remains, however, where the architecture will go after these solutions mature in the market. Allan believes that, for high performance applications, the natural evolution is toward an FPGA-centric system. “By taking advantage of the flexibility and performance of FPGAs,” he explains, “you can create optimized, custom hardware architecture for each application.” Traditional computing starts with a fixed processor architecture and uses compilers to mold the algorithm to the architecture. FPGA-based reconfigurable computing does just the opposite – it molds the processor architecture to the algorithm.
While this observation makes a great deal of intuitive sense, Allan doesn’t believe FPGA-based computing will take over everything. “Von Neumann-influenced machines are a very hardware-efficient way to implement a wide variety of algorithms,” Allan observes. “FPGA-based solutions are like jet engines. While jets have long been known to be much more efficient than piston engines in delivering power, we still don’t drive around with them in our cars. When we really need performance, however, we turn to the technology that can deliver it. It will likely be the same for computers. The everyday, routine computing tasks are handled well by traditional processors, and there will probably not be a motivation to replace them. For high-performance applications, though, we’ll push FPGAs for now.”
When much of the computing world really needs performance, they’ll likely turn to innovators like Allan Cantle and Nallatech. We’ll continue to watch as they develop and deploy dazzling FPGA-based custom computing hardware that will raise the bar on conceivable computing performance. Whatever the high-performance computing architecture of the next decade turns out to be, Allan will be thinking of ways to push the limit. Just like those days when he was 14 with the Computer-in-a-Bag, nothing interests him more.
Allan Cantle is president and CEO of Nallatech.