IBM 1620 REPLACEMENT TRANSISTORS FOR FL/ARW AND WJ CARDS ORIGINALLY USING 086 AND 071, RESPECTIVELY SUMMARY AT THE END Also applies to 1401 AQU card which is similar to ARW. 1410 DFY card: Try MPS651 or STSA1805. Needs to conduct 600mA and block 50V. DAVID WISE MAY-AUGUST 2016 Updated January 2022 When I acquired 1620 #10198 in 1978, I tested every SMS card before ever attempting to run it. I found and replaced about a dozen weak or failed transistors, including one 086 on a type FL Decode Switch card. I did not have a spare 086, or any Germanium transistor that could replace it. Examining the schematic, I understood that Silicon could be used, provided the Base-Emitter junction was protected from reverse breakdown. I removed a wire jumper and installed a 9V zener diode in its place. Electrically, this was in series with the base, with the 083 driver on the left and the new transistor - a 2N2222 - and the pullup resistor on the right. This reduced the Base-Emitter reverse potential to a safe 3V. Initially it appeared to work, but soon I saw that memory performance was poor, with many "dropped" bits. After cringe-inducing modifications to the Sense Amplifiers and exacting adjustment of the drive current, it was reliable - barely. Thirty years later, in 2008, one of the four type 091 transistors in the type FM Current Driver cards failed. It was atypical in that it didn't go open or short, or even lose gain or go leaky. It looked fine on a curve tracer. But it switched off very slowly, which led to a weak and delayed select pulse. As before, I had no spare 091, and no other Ge part even slightly qualified. Eventually I discovered that the General Electric D44C8 Silicon transistor could not be distinguished from the good 091's on the other three FM's. Life was again good - I backed out the sense amp mods, and drive current margin was healthy - but I became curious about my old FL repair, and I discovered that I had overlooked something. The switching cores in the Matrix Switches, and the main memory core array that is their load, are inductive. When faced with a change in current, an inductive load manipulates voltage trying to prevent the change. In our case, the Current Driver shuts off its output current by diverting it to -12V. This reverse-biases the output diodes, disconnecting the load, but not at first. The load follows the 091 collector down, and its stored energy only drains away when the 091 comes to rest at -12V. The nine 086's that are off see the same swing but don't interfere with it because their bases are -12V. My 2N2222's base was only -3V. As soon as the collector dipped beneath that, it started to conduct; in fact, the emitter did too: being positive with respect to base and collector, roles became reversed and the emitter thought it was the collector and vice versa. My transistor was operating in "Inverse Mode", and this unexpected current drain weakened the flip in the matrix switch. I could prevent this with a simple rearrangement, placing the zener in series with the emitter instead of the base. When the switch is on, the zener enters forward conduction like an ordinary PN diode; when the switch is off, the zener enters reverse breakdown and holds emitter at -9V while base is -12V. BREAKING NEWS 2018: I just realized the zener leaks as much as the base, which means I'm _still_ zenering my transistors, albeit slightly. To be perfect, I should install B-E resistors. What a drag. I should have left the zener on B and added a PN diode in the collector like the ARW card. That is elegant electrically, but although the FL card layout facilitates base zener on both channels, it supports collector diode only on one channel. The other requires cutting a trace and putting the diode on the solder side, which won't fit. That disqualifies TO-5 leaving TO-92, TO-251, and TO-126. ----------------------- In "Current_Driver_Candidates", I proposed that the various transistors I came up with to replace the FM card's 091 transistor might also work in the FL Decode Switch and WJ Inhibit Driver cards. I was wrong. The 091 replacements that I tested were too slow. We need smaller, lighter transistors that don't saturate so hard at the base current these cards use. A silicon transistor replacing IBM's 086 transistor in an FL card has its base pulled hard to -12V. To prevent reverse breakdown, add a 9V zener diode in series with the emitter. It must be a type that conducts like a normal diode in the forward direction. It must also have a short reverse recovery (Trr) compared to the transistor, so it reverses before the B-E junction. This will disqualify small, agile transistors with short storage time. Since the collector current is off for microseconds before the base changes state, we measure zener recovery at the transistor base current of 16mA. (We'll assume that if the zener is faster at 16mA, it's faster at the real reverse current, which averages 200mA for type 083 driver.) Before I studied the ARW Decode Switch card, I believed that it's important that the zener be in the emitter lead, not the base, nor replaced by an ordinary diode in the collector. The transistor is operated in "inverse" mode to clamp a negative inductive kick, and this action must begin at 12V, not significantly more or less, to avoid excess noise on the one hand and weakened core flipping on the other. But the ARW card, which is listed as "interchangeable with FL", does precisely that, puts a diode in the collector lead, which means that the FL's -12V clamping action is not important after all, only the fact that it doesn't begin until -12V. Still, the zener is the simplest way to meet the twin objectives of protecting the junction and delaying clamping action. The ARW turns off the base with a parallel RC network, with a diode to protect the base from reverse breakdown. You can drop Silicon straight in, the only constraint being base storage charge: it must be less than 24nC, the charge removed through the capacitor. A silicon transistor replacing 071 in a WJ card does not need protection, since its base sees only -6V. (Current-limited by 140 ohms.) DISQUALIFIED 2N697 Beta at 300mA 2N4401 Ts at 20mA is 50ns same as zener diode MPSW3725 Obsolete - use 2N3725 instead 2N2222A Specs 2N2219A Mechanical 2N3725 Mechanical 2N5320 Mechanical TEST JIG My WJ test circuit is a 2N4401, with base driven by +3V/-7V pulse through 270 ohms. Emitter is at -2.6V. DUT emitter is at 0V, and a tantalum cap between. Driver collector has 15 ohms leading to 160 ohm pullup to +5.5V, and two diodes to DUT base. DUT base has a 470-ohm pulldown to driver emitter. DUT collector has 51 ohm noninductive resistor to +15V. This gives +14mA and -7mA on the base, and 300mA on the collector. <<<< CRAP I AM DRIVING DUT TOO HARD, SHOULD BE 16mA i.e. 3.5V NOT 5V >>>> <<<< MEASURE ALL FL AGAIN TO WINNOW OUT INSUFFICIENT BETA >>>> The FL circuit is similar except no diodes, driver emitter is -3V, and driver pullup is +5V. DUT base is +19mA and -200mA, same as a real FL card with 083 driver. (1.3mA and beta = 150.) Since FL/WJ reverse-bias the base, the breakdown voltage criterion is either BVcbo or punch-through. Except for the rare Ge datasheet that lists punch-through, use BVceo as it's the most conservative. The emitter is at ground for Ge or ARW, -9V for Si in FL, so a 30V spike may be 40V. Your transistor should be at least 10V more; that's 50V for Si/FL. My West-German 2N1304's test good but only because they exceed spec; the punch-through rating is only 20V. ORIGINAL TRANSISTOR CHARACTERISTICS From "Appendix B. Sample Transistor Characteristics" in "Customer Engineering Manual of Instruction - Transistor Component Circuits", Form223-6889-TransistorComponentCircuits.pdf Quote: "These figures are _not_ transistor specifications." They don't say what Vce beta was measured at. It also seems ambiguous to me whether beta was measured at Ic, or Ic is simply the max rating. Type BVcbo@55C BVebo@55C Icbo@55C Ic/Beta/Vbe Turn-on Turn-off 071 20 10 65uA 300mA/25/0.9V 0.65us 083 20 10 36uA 50mA /40/0.45V 0.8us 1.0us 086 35 6 50uA 350mA/32/1.3V 0.8us 1.0us SMS CARD CHARACTERISTICS From "http://bitsavers.trailing-edge.com/pdf/ibm/1401/ce/1401_CE_Drws_1962.pdf" AQU Alloy Switch Decode (071 driving 086) Ton 350ns max Toff 450ns max AQV Alloy Driver Z (071 driving 086) Ton 900ns max Toff 350ns max DKA Current Driver (like FM) - output Ton 200ns max Toff 100ns max READINGS Note: Tf is rise time of collector voltage i.e. fall time of current. In WJ, the silicon parts below can be sped up with a simple Ge Baker clamp like I did on 2N3906's replacing drift transistors back in 1978. Some parts, e.g. KSC1008, don't respond to this treatment due to quasi-saturation. MEASURE AGAIN WJ: PN3569 TN2219A MPSW3725 2N3053 2N1711 BC141 071 MEASURE AGAIN FL: 2N2219A PN2222A PN3569 TN2219A MPSW3725 2N2102 2N3053 TN3019A 2N1711 BC141 BC639 MJD44H11 Sorted by WJ Toff (left columns). * With 1N34A. Type Ts(WJ) Tf(WJ) Ts(FL) Tf(FL) 071 300 120 On a real WJ card 086 140 360 100 140 MPS651* 80 60 KSD1616A* 90 70 BC337* 140 30 2N5320* 120 100 KSC2331* 180 40 MJE182* 150 100 BD139* 150 100 2N2219A 160 60 150 30 NEED TO MEASURE FL 2N2222A 180 60 PN2222A 200 60 140 20 NEED TO MEASURE FL PN3569 240 60 150 20 NEED TO MEASURE 2N4401 280 60 60 20 TN2219A 300 40 140 20 NEED TO MEASURE ----------------------------------- 2SC4614 400 100 360 20 2N1304 40 500 80 160 MPSW3725 550 30 180 10 NEED TO MEASURE BC337 700 50 100 20 2N2102 700 70 160 20 NEED TO MEASURE FL KSC2331 820 40 130 20 MPS651 900 60 160 20 2N3053 950 100 170 20 NEED TO MEASURE FL 2SC5706 1000 70 120 20 TN6715A 1100 50 140 20 TN3019A 1000 150 230 40 NEED TO MEASURE FL KSC1008 1200 50 140 20 2N1711 900 400 160 60 NEED TO MEASURE 2SD1801 1300 70 250 20 BC141-16 1100 500 280 170 NEED TO MEASURE BC639 1500 150 260 40 NEED TO MEASURE FL KSD1616A 1600 100 240 20 STSA1805 3600 200 500 60 MJD44H11 3500 300 700 40 NEED TO MEASURE FL =================================== Sorted by FL Toff (right columns) 086 140 360 100 140 2N4401 280 60 60 20 BC337 700 50 100 20 2SC5706 1000 70 120 20 KSC2331 1100 100 130 20 TN6715A 1100 50 140 20 2N5320 800 60 140 20 KSC1008 1200 50 140 20 TN2219A 300 40 140 20 NEED TO MEASURE PN2222A 200 60 140 20 NEED TO MEASURE FL PN3569 240 60 150 20 NEED TO MEASURE 2SD1801 1300 70 150 20 MPS651 1200 100 160 20 In Use 2N2219A 160 60 150 30 NEED TO MEASURE FL MPSW3725 550 30 180 10 NEED TO MEASURE 2SD1802 190 20 2N1304 40 500 80 160 Voltage, Beta MJE182 1500 100 220 20 ----------------------------------- KSD1616A 1600 100 240 20 BD139 1800 100 240 20 2SD1803 250 20 TN3019A 1000 150 230 40 NEED TO MEASURE FL BC639 1500 150 260 40 NEED TO MEASURE FL 2SD1815 310 20 BC141-16 1100 500 280 170 NEED TO MEASURE FL 2SC4614 400 100 360 20 2SD1815 350 59 STSA1805 3600 200 350 50 MJD44H11 3500 300 700 40 NEED TO MEASURE FL 2SD1816 800 2N3421 900 80 My PN3569, PN2222A, TN2219A, and 2N2219A did okay but the datasheets don't guarantee it. Anyway they're faster than the zener diode, like the 2N4401. For ARW, select for gain at 300mA/1V/cold. Selected 2N130x may work, check gain and voltage. FL Reverse Recovery Time Trr (+20mA, -20mA) Transistor must be longer than zener. ARW Storage Charge Qb (+20mA, -20mA) Transistor must be less than (12V * 2nF = 24nC). Zener : 50ns 2N4401 : 50ns <<< Disqualified for FL BC337 : 600ns = 12nC BD139 : <<