TO: SCSI Parallel Interface Work Group X3T9.2 / 92-219R1 DATE: 12/3/92 RE: Proposal for Hot-Plug Leakage Current Specification By: Tak Asami, Western Digital Corporation 8105 Irvine Center Way, Irvine CA 92718; EMail(Internet): asami@wdc.com =============================================================================== Abstract: ========= SCSI-3 Parallel Interface document now defines the hot-plugging procedures (See X3T9.2/91-010R7 Section 7.5). Up to now, supporting this feature involved the design of drivers that do not disrupt the bus during the power-up. But as new devices are designed in sub-micron process, a new element emerged as a potential hazard. The protection against electro static discharge (ESD) is typically process specific and the finer the process, more difficult it becomes. Several new implementations of ESD are introduced for sub-micron processes, and some of them are found to draw large amount of current under certain circumstances that may occur during hot plugging process. I would like to ask the work group to incorporate a separate specification for hot-plug leakage current to the standard, so (a) silicon vendors have practical design goals for the functional hot-plugging, and (b) we can make a value judgement on what implementation to adopt. Issues: ======= SPI document dated November 10, 1992 has a section describing the hot-plugging procedure (Sec 7.5 "Removal and Insertion of SCSI devices"). CASE 4 "Current I/O process allowed during insertion or removal" is the most critical. According to this, the insertion / removal of the device is done with its power off. This means the power to the device (VDD) is at 0V, while there are negation level voltages (> 2.7V) on several pins. Because of the way the ESD circuit works, this may trigger the protection device and creates a short to either power or ground node. This short circuit typically lasts only until the internal voltage reference is charged up to a certain level, which may take several micro seconds. Therefore, when the device is physically inserted to the SCSI bus, large current surge occurs (we simulated one case as large as 400mA peak for 400nsec). This makes hot-plugging impossible. All of this is independent of the transceiver circuit; it can happen even though the driver itself is shut off during the power up/down. And it is expected only at physical insertion. And it is only expected to be tran- sient, i.e., it will die down eventually. There is no surge current expected during removal, as well as power up / down associated with the ESD protection. The driver should be designed to stay off during power up / down. There are a few ways known to implement the ESD protection without suffering such consequence, but most are patented or patent pending. This may make for more expensive protocol devices and/or they may be available from only limited number of sources. I do not believe it is in the standard's place to require using a certain technology, especially if it is a proprietary technique. Rather, it should provide for how much disturbance the bus can tolerate so silicon vendor can work to optimize the implementation. In that spirit, I propose the following changes to the SPI document. Proposed Change for SPI ======================= Following is the section of the current SPI document (X3T9.2/91-010R7) that I would like to see changed; underlined lines are added, and overstriked lines are to be deleted. 7.1.3. Single Ended Input Characteristics SCSI devices with power on shall meet the following electrical characteristics on each signal (including both receivers and passive drivers): a) VIL (Low-level input voltage) = 0.0 to 0.8V dc b) VIH (High-level input voltage) = 2.0 to 5.25V dc c) IIL.ON (Power On Low-level input current) = 20uA at Vi=0.5V d) IIH.ON (Power On High-level input current) = 20uA at Vi=2.7V e) Minimum input hysteresis = 0.3V dc In addition, transient leakage might be expected at the time of physical ------------------------------------------------------------------------ insertion of a SCSI device. It is expected to be exponentially decaying ----------------------------------------------------------------------- current, that does not exceed the following specifications. ----------------------------------------------------------- f) IIH.HP (Hot-Plug High-level input current peak value) = 1.5mA at Vi=2.7V -------------------------------------------------------------------------------- g) THP (Transient current duration to 10% of peak value) = 20usec max -------------------------------------------------------------------------- It is recommended that SCSI devices with power off also meet the above IIL and IIH electrical characteristics on each signal, except at the time of physical ------------------------------ insertion, when IIL/H.HP and THP prevail. ----------------------------------------- (No more changes to the remainder of the section). 7.5. Removal and insertion of SCSI devices This clause defines the requirements for removal and insertion of SCSI devices on the SCSI bus......... Four cases are discussed. The cases are differentiated by the state the SCSI bus is in when the removal or insertion occurs. In all of the cases, only one ----------------------------- SCSI device is inserted or removed at a time. --------------------------------------------- Justification: ============== The justification for the IIH.HP is as follows: The SPI document specifies the maximum impedance for the cable to be 96 ohm. In a transmission line, the instantaneous voltage fluctuation due to device ` insertion is governed by the amount of current and the cable impedance. In other words, a current can cause only so much voltage fluctuation as the cable impedance permits, until the wavefront reaches the terminator at the end of the cable. Assuming a system can tolerate 300mV (minimum hysteresis) of sudden fluctuations, the maximum leakage current must be less than: IIH(max) = 300mV / 96ohm = 3.13mA And the transient current can be expressed approximately by: iIH(t) = IIH * exp(-t/tau) tau is a time constant of this leakage function, which typically in the order of a few msec. Once the wavefront reaches the terminator, more current is available to absorb the disturbance. Considering a physical limitation where only one device can be inserted at a time (which is described in the proposed revision to the section 7.5), if the peak leakage is restricted to this value, which decays exponentially, no harm should come to the system. By further guard-banding the specification by 100%, the peak value should be limited to 1.5mA. Note the proposal only specifies the high level input current; this is because the voltage fluctuations are expected only when the bus signal level is at high level. There is very negligible disturbance when the level is low. If the "insertion" occurs exactly at the signal transition point, the built-in hysteresis of the receiver should protect it. Notice also that no disturbance is expected when the device is connected to the bus but not powered on, due to the transient nature of this phenomenon. As long as the driver is designed to remain high impedance when the power is off, there should be no concern to this effect. --- Insert "Sample Waveform" here --- Measurement Method: =================== The transient leakage current discussed here can be measured using a simple setup. Using a general purpose pulse generator, a 2.7V square wave is injected to a SCSI pin (all other pins are open), and measure the current through the VSS pin. VDD pin is left open. Every rising edge represents the "physical insertion". --- Insert "Measurement System Diagram" here --- Summary: ======== o Up to 1.5mA of transient leakage should be allowed during the physical insertion of a SCSI device. o This current may be measured using a simple set up. o No ESD protection or any other circuit that incur significant DC leakage should be used. o This hot plug current specification does not in any way, restrict the SCSI devices from being connected to the bus yet powered down. o Only one SCSI device can be inserted / removed at a time. o This specification is independent from Sec 7.1.2. where it specifies "All single ended drivers shall maintain the high-impedance state during power-up and power-down cycles....". The driver leakage and power up/down cycle are not what this proposal is intending to cover. X3T9.2 / 92-219R1