A blade server is essentially a single circuit board populated with components such as processors, memory, and network connections that are usually found on multiple boards. The explosive growth of mobile applications and the increasing use of Software as a Service (SaaS) has made protecting the blade servers that store data “in the cloud” and keeping them operating efficiently more important than ever. Making that happen depends in part on preventing over-current damage that can shorten their lives and disrupt datacenter operations.
One factor complicating ensuring blade protection is the trend toward concentrating ever-higher levels of computing power on each blade. Once, blades typically ran at maximum power levels of 200–400W; today, however, blade power levels of 600–800W, sometimes even 1000W, are increasingly common. The trend toward increasing power has developed at the same time as a trend has emerged toward decreasing blade operating voltages in order to reduce DC-DC conversion costs and power efficiency losses within the data center power distribution scheme. In some popular designs, a typical low side voltage is 12VDC, which can drive input current levels to greater than 80A per blade. For the blade’s circuit designers, the challenge is to ensure over-current fault protection in a cost-effective, reliable, safe and feasible way to prevent costly hardware damage and service outages.
Because blades are intended to be hot-swappable (i.e., to allow maintenance personnel to remove them from the chassis and plug in new ones while the main power is still on), they are designed with a hot-swap controller IC. This IC provides the primary fault protection mechanism by first sensing over-current and other fault conditions, then signaling a MOSFET to shut down power to the board. The objective is to prevent adjacent blades within a chassis from going down due to undervoltage lock-out (a short circuit can cause the backplane power rail voltage to go down due to droop) and also prevent damage to the faulted blade and backplane. Critical components within this system are the backplane traces, backplane connectors, and power connector on the board. Backplane damage can be particularly problematic because it could lead to the loss of an entire chassis within the data center.
Secondary protection with physical fuses is necessary to back up the primary hot-swap controller IC. These physical fuses provide independent over-current fault protection as a backup to the main IC in case of damage to the MOSFET or IC during fault conditions. To maximize the number of blade servers that will fit into a standard 19-inch chassis, blades must be as thin as possible, so the use of low profile surface mount fuses is essential. However, no manufacturer currently produces a 60A- or 80A-capable surface mount electronic fuse. To support the blade’s full input current, a blade’s circuit protection designer must parallel two or more fuses. However, doing so safely can be complicated for very high current applications.
The highest current SMT fuse now on the market is rated for 40A. To take proper fuse de-rating calculations and the de-rating necessary when paralleling fuses into account, a designer would need about 150A of total rated current from the fuses or four 40A fuses in parallel. In addition, circuit designers must be concerned with how the fuses will behave and carry current between them because slight differences in fuse resistance will generate different carrying currents on each fuse. They also need to understand how the system of fuses they’ve created will react during a short-circuit event. To maintain high system reliability, it’s critical that the physical fuses do not trip before the hot-swap controller IC has the time needed to remove power from the board.
Circuit protection designers can parallel multiple fuses safely to increase a blade’s current carrying capability if they understand these points:
·All the fuses in parallel must have the same fuse rating. If possible, all the fuses should be from the same production batch to be reasonably certain the DC resistances of the elements of the fuses match closely.
·The maximum interrupting current or voltage is equal to that of an individual fuse. In other words, the maximum interrupting current or voltage for the parallel combination is equal to that of one individual fuse only, and is not equal to the sum of the interrupting currents or voltages of all the fuses.
·All the fuses working together will attempt to share the load current equally by increasing the resistance of a fuse element as more current attempts to flow through it. This increase in resistance is caused by the temperature increase of the element, which was caused by the current increase. Current unbalance is consequently minimized in a parallel configuration and the system will be self-regulating. Excess current in one fuse will be shed to its neighbor and a state of equilibrium will soon be attained. Because there will never be a perfect 50/50 or 25/25/25/25 current split, Littefuse recommends applying a 20% derating factor to the combination.
·Close thermal tracking is required to keep the fuses at the same temperature, with respect to both ambient temperature and normal operating temperature. Ensure all fuses are exposed to the same airflow and have similar heat conduction mechanisms acting on the wire leads or fuse clips.
·Always remember that the melting integral (I²t) will increase by the square of the number of fuses in the parallel combination. For example, if two fuses with an I²t = 3.5A²s are used in parallel, the effective I²t will be 3.5A²s x 2² = 14A²s. If three are used, I²t = 3.5A²s x 3² = 31.5A²s.
To learn more about choosing and using fuses safely for high current circuit protection applications, download a free copy of Fuseology: Fuse Characteristics, Terms and Consideration Factors from Littelfuse.
Bharat Shenoy joined Littelfuse in 2008 and is currently Director of Technical Marketing. His career in the circuit protection industry began in 2001. He has served in various applications engineering and sales management roles. Bharat has a B.S. in Naval Architecture from the United States Naval Academy and served seven years as a Naval Officer in the Nuclear Submarine fleet. Bharat is married with two sons and lives in San Jose, California.