09 August 2011
Steve Edwards, cto, Curtiss-Wright Controls
Graham Pitcher finds out that COTS is now being applied at the system level.
The world of military systems procurement has changed enormously. Gone are the legendary 'costs plus' contracts and proprietary systems, replaced by a sharp focus on the use of commercially available technology.
"The market is changing," said Steve Edwards, chief technology officer for Curtiss-Wright Controls Embedded Computing business unit (CWE). "Consolidation means it is now more cost effective for major primes to outsource and this has brought Curtiss-Wright higher up the 'food chain'."
CWE is focused almost entirely on the mil/aero market. "Although we have some commercial sales," Edwards noted, "our focus is on deploying technology into harsh environments; and the mil/aero market is demanding."
The company is a COTS vendor of open architecture products, including VME and VPX based devices, as well as some CompactPCI. It also develops and sells mezzanine boards. "Everything we design is done under our own steam," Edwards asserted. "We maintain our own road map and have an active product development programme. This means that 90% of what we sell is developed internally, with the remaining 10% contributed by NRE based work, generally modifications to existing products."
Edwards sees a number of trends at the high performance end of the mil/aero market. "With big radar projects, and with intelligence, surveillance and reconnaissance (ISR) in general, developers are trying to take in massive amounts of data and do the processing locally. They have to do the processing in the same location because they simply can't send that amount of data over an rf link. So commercial technology is being used more and more."
A consequence is that time to market schedules found in the comms world are now being adopted in military projects. "These used to have a lead time of five to seven years," Edwards noted, "but now it's two to three years."
He also sees the adoption of Intel processors for high performance, including the Core i7. Alongside this is a move to the use of general purpose gpus, or gpgpus, for high end signal processing. "These aren't good for everything," Edwards accepted, "but there are some problems where they provide an order of magnitude performance benefit. This is generating a lot of interest from our customers."
Because of its strong interest in integration, CWE is a supporter of open architectures – it has recently joined the Open Fabrics Alliance. "Something like an OpenVPX backplane allows different technologies to be combined," he continued, "and this is attractive. It's certainly the way high end projects are going, where developers want to use standard hardware and standard middleware in order to make sure the system isn't proprietary."
By necessity, CWE has to work closely with its suppliers. "We work with key vendors – such as Intel, Freescale and Nvidia – to make sure we have a road map to buy parts for up to seven years. And Intel, for example, has a group that deals with taking products from other areas and configuring them for the embedded market. We need to ensure that, in six months' time, parts don't disappear."
With gpgpus, CWE plans to take a module based approach. "We're looking at modules commonly used in laptops," Edwards explained. "If Nvidia brings out a newer module, we'll have a pin compatible device with similar functionality and power consumption."
Nevertheless, CWE still has to support programmes for up to 20 years in some cases. "But various government agencies are realising that, as they adopt more commercial technology, they have to have a more commercial approach to technology refresh. While they are struggling now," Edwards observed, "we'll see more frequent refreshes in the future."
Partly, this will be accomplished by swapping out elements of a system. "If there is an open architecture backplane and we have a next generation processor to fit the slot, we can offer an upgrade," he said.
Edwards noted that CWE is 'aggressive in introducing products to market'. "We pride ourselves in getting products to market only a few months after a chip we're committed to samples. If it's key to our business, we'll come out with a next generation product."
Modules are also important, particularly when it comes to the application of fpgas. "Modules are critical to the strategy," he asserted. "FPGA development is complex and takes longer than other products. And new generations of fpga don't come as quickly as, say, a cpu from Intel. While fpgas don't change that quickly, I/O requirements do. Using an FMC module means we can spin new analogue front ends or digital I/O in a couple of months, usually at relatively low cost. It helps us because we don't have to spin a new base card at several million dollars."
There's also a trend towards smaller systems, which are often space constrained. "Small ISR systems are now in 3U housings; sometimes smaller," Edwards said. "Now, developers have the ability to move what were systems to 6U cards, and what was on a 6U card to a 3U card. Processors are far more powerful, which means a lot more can be done in smaller systems. We can build what used to take several boards on a 3U card today."
As COTS becomes a way of life at the board level for mil/aero companies, attention is now moving to sub systems and even systems. "Open architectures are increasing compatibility at the sub system level," Edwards concluded, "and proprietary interfaces are being frowned on, even at the chassis level."
Steve Edwards joined Curtiss-Wright Controls in 1998 as a senior hardware engineer and has since held numerous positions, including technical product lead and product development manager.
More recently, as technical product manager, Edwards was responsible for product definition, architecture, technical bid response and pre sales customer interaction on major programmes.
He developed Curtiss-Wright Controls' first fpga based computing platform and was named a Technical Fellow of the company in 2008.