VideoJohn Chambers likened it to a 100-year flood, although the problem was dearth, not plenitude. The swift evaporation of technology demand that began in the latter part of 2000 was indeed exceptional, as the CEO of Cisco Systems famously suggested. Chipmakers and PC companies suddenly found themselves with a glut of inventory and capacity. Networking and telecom equipment makers were particularly hard hit; Cisco, more irrationally exuberant than most, was forced to write off a staggering $2.25 billion worth of gear. Throughout the first half of 2001, a procession of high-tech companies — including such bellwethers as Nortel Networks, Lucent Technologies, Corning, and JDS Uniphase — announced huge write-downs of unsalable inventory.
Today, high-tech companies are still loaded with rapidly depreciating goods. At one end of the food chain, the cyclical semiconductor industry is suffering through its deepest trough in demand since 1998, the year of the Asian crisis. In the middle, electronics contract manufacturers and their suppliers, customers, and distributors are trying to figure out who owns which surplus components. At the other end of the chain, PC makers are waging price wars, and the gray market in networking equipment is thriving.
Flood of the century or not, tech companies are taking steps to limit their exposure to the next traumatic event. Some are revising their inventory models; others are implementing supply chain software and setting up Web supplier hubs. Everyone wants tighter collaboration with suppliers and timelier information from customers. Tech companies are trying, in short, to make their supply chains shorter, transparent, and as flexible as possible.
New Logic
Check out the recent earnings releases of semiconductor makers (not the pro forma kind) and you'll find a litany of inventory write-downs: Agere Systems, $270 million; Micron Technology, $260 million; Vitesse Semiconductor, $50.6 million; Alliance Semiconductor, $50 million; Xilinx, $32 million. Worldwide, chip sales in June were down 30.7 percent from a year ago, according to the Semiconductor Industry Association, and analysts predict a decline in 2001 revenues of more than 20 percent — the steepest ever.
"I've been in the chip industry for 20 years," says Nathan Sarkisian, "and I've never seen anything like it." Sarkisian is senior vice president and CFO of Altera Corp., a San Jose, California-based chipmaker with 2000 revenues of $1.4 billion. "We grew roughly 65 percent last year with less than four months' supply of inventory throughout most of the year," he recalls. "That's pretty good when you think about semiconductor product cycles."
But in the fatal fourth quarter, units shipped to distributors fell 25 percent short of expectations. The slide continued into 2001, thanks to declining demand from Altera's major customers, communications companies. For Q2 2001, revenues were down 25 percent sequentially and 37 percent from Q2 2000. Altera was eventually forced to write down a whopping $115 million worth of inventory.
Going forward, Altera wants to ensure that future market dips won't savage the bottom line, and to that end it's revising its inventory model, for starters.
Altera designs programmable logic devices (PLDs). It's a "fabless" chipmaker, outsourcing manufacturing to giant foundry Taiwan Semiconductor Manufacturing Corp. Previously, it would build its mainstream PLDs through to finished goods, stockpiling them in Asian facilities in anticipation of customer demand. "We own the inventory as soon as it leaves the fab," notes Sarkisian. Also, it would essentially build new products on spec, producing quantities well beyond what the customer needed for prototyping. The virtues of this model are highlighted in Altera's annual report: "We, our distributors and subcontract manufacturers — not our customers — hold stocks of inventory, thereby enhancing the cost advantage of PLDs for our customers."
Now, Altera will continue to build its mainstream products to stock, but only in die banks (stores of chips before packaging and testing). "By building die, we have taken out the biggest portion of the manufacturing lead time, but the inventory is in its most flexible form, with a minimum of value added," says Sarkisian. Only when orders are confirmed will Altera's subcontractors package, test, and ship the PLDs.
The lead time for these products will be measured in weeks. For Altera's mature products, "we will be strictly build-to-order," says Sarkisian, and the lead time for those will be measured in months. Finally, new products will no longer be built on spec; a customer order will be required.
Visible Improvements
Chipmakers are at the mercy of the laws of physics. It takes anywhere from three to seven weeks to turn a raw silicon disk into a wafer with hundreds of chips, depending on the complexity of the chip and how much a customer is willing to pay, says Jim Kupec, president of United Microelectronics Corp. USA, a division of Taiwan-based foundry UMC. Additional time is required to separate, package, and test the chips. And in the real world, "things get spoiled in the fab," says Arnold Maltz, associate professor of supply chain management at Arizona State University College of Business. "Every now and then, somebody brings the wrong batch. Capacity isn't always available. Then you have the mismatch of supply and demand." In a 1999 study of major U.S. chipmakers, Maltz and his fellow researchers found that the average cycle time for semiconductors, from the fab to the customer, was 117 days — plenty of time for demand to change direction.
Reader Comments» Post a comment