Long-time ICE supplier Applied Microsystems (AMC) is getting out of the embedded systems tool business.
The company was one of the first to provide third-party in-circuit emulators. They were one of the best and for many years one of the most dominant forces in the embedded tool biz. But the market has changed, demand for their products slowed, and their stock price has collapsed, declining to 20 cents per share from the $5 to over $20 they commanded in the '90s.
I remember seeing AMC's products at shows in the early '80s. Their first ICEs were small boxes with hex keyboards that offered little connection to a host computer. This is before the PC, before universal desktop machines. The 4 MHz speeds and minimal complexity of the 8 bit embedded processors of the time demanded little of the tools. As I recall, these units sold for about $3k.
Times changed. Embedded systems grew up, many moving from slow 8 bitters to high performance 16- and 32-bit processors. C replaced assembly. Prefetchers, pipelines, cache, and other performance-enhancers stretched the capabilities of tools. Applied rode the wave, successfully adapting their product line to the changing realities of the market.
When I started my own ICE business in the '80s I foolishly believed that keeping prices low would lead to success. Wrong. I learned that customer support is the most expensive of all components in an ICE. A $600 device or a $6000 unit had about the same support needs, so how can one afford to sell cheap? Applied, too, ratcheted their sell prices up, by the early '90s pushing units to the $20k+ range.
That's a buncha bucks for a debugger and a hard sell to the boss. “You want 20 grand for a device to find your mistakes? Just don't make any!” Though my sympathies are entirely with Applied, customers complained and protested, but had few alternatives.
Then Motorola invented the BDM (background debug mode), which put most of the logic needed for a debugger on-chip. Nowadays many embedded processors include these circuits. For the cost of a few ICs and a software debugger the developer gets a pretty nice development environment. It's not as sophisticated as one finds in an ICE, but BDMs offer an almost unbeatable value. Support is much cheaper since BDMs use a very simple connection to the target system, one that doesn't need 100 perfect high-speed signal paths.
Applied responded to the changing market with new mixes of products. But even sales of thousands of BDM tools could never make up the company's $40 million yearly volume. Scaling down ICE technology to lower-cost units like their CodeTap and SuperTap helped, but not enough. Their very cool CodeTEST, sort of a debugging adjunct that trapped memory leaks, collected deep traces, and measured code coverage just did not catch on as it could or should have. I figured that the code coverage requirements in DO-178B ensured success of that tool, but customers were scarce. Betcha the tool was ahead of its time, and that an increasing awareness of safety-critical issues will spawn more, not fewer, needs for this sort of product.
Today AMC's sales have crashed to half last year's levels and the company is furiously bleeding money. Subject to regulatory and stockholder approval, AMC will sell their tools off to Metrowerks, a long-time player in the software end of the embedded world. Metrowerks is getting a great deal, acquiring Applied's 20 years of R&D for less than $4 million.
The remnants of Applied will leave the embedded space and hopefully thrive in the “easier” world of the Internet servers.
I wish both companies well, and sure hope Metrowerks finds a way to profitably provide AMC's and other tools to the embedded community. We need all of the debugging help we can get! I suspect, though, that AMC's troubles stem from fundamental changes in this market. More developers than ever are very removed from the hardware. More organizations segregate low-level hardware/software whizzes from applications people. Most of the latter have no need for expensive but powerful debuggers. A BDM provides all of the power needed for app-level troubleshooting.
Are all of the tool companies in trouble? Green Hills sure seems to be an exception, recording strong growth and profits. They advertise constantly. But looking through this month's Embedded Systems Programming I see no ads from Applied. None from Metrowerks. Avocet (who bought Huntsville, Softaid, Cactus, and 2500AD): nada. Nor any from Wind River (who bought, well, everyone). Red Hat appears to be backing a bit away from eCos — and they're not in ESP either.
SEC filings from both Wind River and Applied both blame the collapse of telecom on their travails. Surely there's a lot more to this industry than cell phones and the Internet!
What do you think? Is the era of hardware debuggers dead?
Jack G. Ganssle is a lecturer and consultant on embedded development issues. He conducts seminars on embedded systems and helps companies with their embedded challenges. He founded two companies specializing in embedded systems. Contact him at . His website is .
With the last two projects being BGA packaged system on chip at over 200 MHZ with multiple processors in one chip, and C++ being used for development,traditional emulators lacked the ability to physically probe the parts. (A 1 foot ICE cable would make a 20% timing error at 200 MHZ)
JTAG and BDM emulators were usefull for the low level de-bugging of the OS, but were limited in source level debugging of C++.
Often source debuggers now connect via ethernet or high speed serial link and are largely a software item that depends partly on some specialhardware on the SOC. (Lack of enough of this hardware delayed the program by 1 year)
On the last program that used a traditional ICE, it was saved for the really hard problems, and the bulk of the debugging was done using aPC to drive the target hardware via a bus cycle simulator. This way everyonewas able to use visual C++ to debug in a full PC environment — much more productive and cost effective than the ICE for most of the code.
(In Summary)The ICE is getting harder and harder to hook up to a fast processor, and Pre-built RTOS's, Protocol Stacks, etc. mean less need for an ICE to solve the really hard problems — Modular Lego building block design (to use up all those cheap mips) as opposed to the old hand crafted approach.