The TO-66 package is a hermetic two-lead metal can built on a diamond-shaped mounting flange, registered by JEDEC as TO-213AA, with the case acting as the collector. Devices in it dissipate anywhere from 20 W to 90 W at a case temperature of 25 °C, depending on the die inside. It bolts down with two 4-40 screws. Today the outline survives mainly as qualified military inventory priced between $10 and $50 per device.
Key takeaways
- TO-66 is now formally TO-213AA. TO-123 and TO-124 were re-registered as TO-213AB and TO-213AC. All three share the 0.200 in (5.08 mm) pin spacing and the same mounting footprint.
- The outline tells you nothing about the rating. Verified parts in this package span 20 W to 90 W and 1.94 °C/W to 7.5 °C/W junction-to-case.
- Interface thermal resistance for a TO-66 flange at 6 in-lb is 0.5 °C/W metal to metal with grease, and 0.9 °C/W through 2 mil mica with grease (onsemi AN1040/D, Table 1).
- Free air is not an option. The 2N3584 and 2N3585 are rated 35 W at TC = 25 °C and only 2.5 W at TA = 25 °C, a ratio of 14 to 1.
- Most of the commercial catalog is discontinued. What remains stocked is Microchip JAN, JANTX and JANTXV product at $28 to $49 each, plus a few Central Semiconductor lines on 2 to 6 week lead times.
- Specify TO-66 for hermetic or MIL-PRF-19500 sustainment work. New designs below 50 W belong in TO-220 or TO-247.
What the TO-66 package is
The TO-66 package is a hermetically sealed metal transistor outline built on a diamond-shaped flange with two mounting holes. Two leads pass through individual glass-to-metal seals at 0.200 in spacing, and the metal case forms the third terminal, normally the collector. JEDEC now registers the outline as TO-213AA.
JEDEC folded the flange-mounted headers with 0.200 in pin spacing into one TO-213 family in Publication 95 in September 1976. TO-66 became TO-213AA, TO-123 became TO-213AB, TO-124 became TO-213AC. The three differ only in flange thickness and overall height. TO-213AB caps the base plate at 1.02 mm (0.040 in), TO-213AA at 1.90 mm (0.075 in), TO-213AC at 2.59 mm (0.102 in), with TO-213AC also about 0.4 mm taller. Holes and pins line up across all three, so they interchange in an existing chassis.
Vendors still print the old name. Microsemi heads its drawing TO-213AA (TO-66) in datasheet T4-LDS-0283; Central Semiconductor labels the identical outline TO-66 CASE. Treat the two designations as one part number space when you search distributor stock.
Construction matters more than nomenclature here. The die sits directly on the flange and the cap is welded over it, so heat leaves through the mounting surface with no plastic body or lead frame in the path. That is also why the flange is live: on a bipolar part it carries the collector, and any heatsink you bolt it to floats at collector potential unless the joint is insulated.
Two-lead versions dominate, with base and emitter brought out and the case as collector. Three-lead and four-lead variants of the same footprint exist, so confirm the terminal count on the specific drawing before laying out a socket.
TO-66 package dimensions and footprint
The figures below come from Microsemi drawing T4-LDS-0283 Rev. 1, cross-checked against EESemi typical-properties data and against the JEDEC TO-213AA limits where the two overlap.
| Feature | Min | Max | Source and drawing reference |
| Cap diameter | 11.94 mm (0.470 in) | 12.70 mm (0.500 in) | LDS-0283 dim A1; EESemi quotes 12.4 mm typical |
| Overall height | 6.35 mm (0.250 in) | 8.64 mm (0.340 in) | LDS-0283 dim N; matches the TO-213AA 0.340 in ceiling |
| Base plate thickness | 1.27 mm (0.050 in) | 1.91 mm (0.075 in) | LDS-0283 dim B; the 0.075 in max is what separates TO-213AA from AB and AC |
| Lead spacing | 4.83 mm (0.190 in) | 5.33 mm (0.210 in) | LDS-0283 dim K; 0.200 in nominal, the family-defining dimension |
| Lead diameter | 0.71 mm (0.028 in) | 0.86 mm (0.034 in) | LDS-0283 dim E |
| Lead length below flange | 9.14 mm (0.360 in) | not specified | LDS-0283 dim D; EESemi quotes 9.1 mm typical |
| Device mass | 5.7 g (Microsemi) | 5.84 g ±10% (Central) | Microsemi LDS-0283; Central TO-66 material composition sheet |
Table 1. Verified TO-66 (TO-213AA) outline dimensions.
One trap is worth naming, because none of the reference pages flag it. Published flange sizes disagree. JEDEC-derived reference data gives base plate diagonals of 31.4 mm and 19.0 mm with the two mounting hole centers about 23 mm (0.91 in) apart on the long diagonal. EESemi quotes a typical base of 34 mm by 20 mm. The gap is far wider than any tolerance band, and the replacement mica washers sold for TO-66 measure 32.5 mm by 20 mm, sitting between the two. Take the footprint from the drawing of the part you are actually buying, not from a generic outline page.
Materials are worth knowing before you write a plating or corrosion requirement. Central Semiconductor reports the TO-66 header as an iron-nickel alloy making up 83.4 percent of device mass, with the can another 15.5 percent, matte tin plating, and a high-temperature lead-bearing die attach. Microsemi specifies nickel plate with a nickel cap and terminals solder-dipped Sn63/Pb37 over nickel-plated Alloy 52.
TO-66 against TO-3, TO-126 and TO-220
Mechanical size is the obvious difference. The thermal interface is the one that shows up in your junction temperature budget.
| Parameter | TO-66 (TO-213AA) | TO-3 (TO-204AA) | TO-126 (TO-225AA) | TO-220AB |
| Body | Hermetic metal, diamond flange | Hermetic metal, diamond flange | Plastic | Plastic |
| Leads out of body | 2 (case is the third) | 2 (case is the third) | 3 | 3 |
| Base plate long diagonal | 31.4 mm | 40.13 mm | not applicable | not applicable |
| Mounting hole centers | approx. 23 mm (0.91 in) | 30.15 mm (1.187 in) | single hole | single hole |
| Max overall height | 8.63 mm | 11.43 mm | not applicable | not applicable |
| AN1040 test torque | 6 in-lb | 6 in-lb | 6 in-lb | 8 in-lb |
| Interface, metal to metal, dry | 1.5 °C/W | 0.5 °C/W | 2.0 °C/W | 1.2 °C/W |
| Interface, metal to metal, greased | 0.5 °C/W | 0.1 °C/W | 1.3 °C/W | 1.0 °C/W |
| Interface, mica, dry | 2.3 °C/W (2 mil) | 1.3 °C/W (3 mil) | 4.3 °C/W (2 mil) | 3.4 °C/W (2 mil) |
| Interface, mica, greased | 0.9 °C/W (2 mil) | 0.36 °C/W (3 mil) | 3.3 °C/W (2 mil) | 1.6 °C/W (2 mil) |
Table 2. Interface thermal resistance rows are from onsemi AN1040/D Rev. 5, Table 1, measured in the onsemi applications laboratory.
Read the last four rows against the junction-to-case figures in Table 3 before you conclude anything. A greased mica joint adds 0.9 °C/W under a TO-66 whose own junction-to-case resistance is typically 5 °C/W, so the insulator costs roughly a fifth of the package budget. Under a 2N3055 in TO-3, junction-to-case is 1.5 °C/W per the STMicroelectronics datasheet, and a greased 3 mil mica joint adds 0.36 °C/W, roughly a quarter. The interface is not negligible in either case, but it dominates in neither.
Running the joint dry is what actually costs you. The TO-66 insulated interface goes from 0.9 °C/W to 2.3 °C/W with no grease, an extra 1.4 °C/W. At 6 W that is 8.4 °C of junction temperature given away for nothing. AN1040 puts it plainly: with mica washers, grease is effectively mandatory.
What a TO-66 device can actually dissipate
| Part | Type | VCEO | IC | PD at TC 25 °C | θJC | Source |
| 2N3054 | NPN | 55 V | 4.0 A | 25 W | 7.0 °C/W | Central R3, 29 Nov 2022 |
| 2N3054A | NPN | 55 V | 4.0 A | 75 W | 2.33 °C/W | Central R3, 29 Nov 2022 |
| 2N3738 | NPN | 225 V | 1.0 A | 20 W | 7.5 °C/W | Central R1, 2 Sep 2014 |
| 2N3739 | NPN | 300 V | 1.0 A | 20 W | 7.5 °C/W | Central R1, 2 Sep 2014 |
| 2N3584 | NPN | 250 V | 2.0 A | 35 W (2.5 W at TA 25 °C) | 5.0 °C/W | MIL-PRF-19500/384 |
| 2N3585 | NPN | 300 V | 2.0 A | 35 W (2.5 W at TA 25 °C) | 5.0 °C/W | MIL-PRF-19500/384 |
| 2N6317 | PNP | 60 V | 7.0 A | 90 W | 1.94 °C/W | Microsemi LDS-0283, from 0.515 W/°C derating |
| 2N6318 | PNP | 80 V | 7.0 A | 90 W | 1.94 °C/W | Microsemi LDS-0283, from 0.515 W/°C derating |
Table 3. Ratings for real devices sharing the identical TO-66 outline. All are rated TJ max 200 °C.
The spread is the point. Same flange, same footprint, same mica washer, and a factor of 4.5 in power rating with a factor of 3.9 in junction-to-case resistance. The 2N3054 and 2N3054A make the case cleanly: identical package, identical pinout, 25 W against 75 W, 7.0 °C/W against 2.33 °C/W. Nothing about the outline sets the rating. Only the die and the die attach do. Any heatsink sized from the phrase “it is a TO-66” is sized from nothing.
The free-air number deserves equal attention. MIL-PRF-19500/384 rates the 2N3584 and 2N3585 at 35 W with the case held at 25 °C and 2.5 W with only ambient air at 25 °C, derating 200 mW/°C on case temperature and 14.85 mW/°C on ambient. That implies a junction-to-ambient resistance near 67 °C/W. A TO-66 with nothing bolted to it is a 2 W part.
Worked example: heatsinking a 2N3585 at 6 W
Take a 2N3585 running 6 W in a 50 °C enclosure, isolated from the chassis with 2 mil mica and grease, on a Thermalloy THM6019 diamond heatsink rated 8 °C/W in natural convection and 4 °C/W at 100 LFM.
Natural convection: TJ = 50 + 6 x (5.0 + 0.9 + 8.0) = 133 °C.
That passes the 200 °C datasheet limit with room to spare, and fails any serious reliability rule. AN1040 cites field data showing failure rates roughly halving when junction temperature drops from 160 °C to 135 °C, and notes that military power supply guidance caps junctions at 110 °C.
To reach 110 °C at 6 W the total path must come in under 10 °C/W, which leaves 4.1 °C/W for the heatsink after the package and the insulated joint are paid for. The same THM6019 under 100 LFM of forced air delivers 4 °C/W.
Forced air at 100 LFM: TJ = 50 + 6 x (5.0 + 0.9 + 4.0) = 109 °C.
One fan changed the answer, and the package contributed exactly half the thermal path in the passive case. This is the calculation that decides whether TO-66 stays in the design.
Mounting a TO-66 without wrecking it
Aavid Thermalloy TO-66 mounting kit No. 4813 defines the hardware set: one Thermalfilm insulator (part 43-66-2), two shoulder washers 7721-6PPS, two number 6 flat washers, two number 6 lock washers, two 4-40 UNC-2A by 1/2 in pan head screws, two 4-40 hex nuts, and one solder lug. Suffix M swaps the insulator for mica, suffix S for Thermalsil III.
Note the screw size. TO-66 takes 4-40 hardware, not the 6-32 most engineers reach for on a TO-220 tab. AN1040 characterizes the flange at 6 in-lb, which is the number to design the assembly torque around.
Five practices carry most of the risk:
- Prepare the surface. Flatness under 4 mils per inch and a finish of 50 µin to 60 µin is enough. Deburr drilled holes, because burrs both reduce contact area and puncture insulators.
- Grease the joint whenever an insulator is present. Thermal compound has a resistivity near 60 °C/W/in against 1200 °C/W/in for the air it displaces.
- Tighten the two screws alternately, finger tight first, then to final torque in at least two passes. Pulling one side down before the other is how a flange gets bowed.
- Never bend the leads at the body. The pins exit through glass-to-metal seals that are not designed to take bending or axial load, and a cracked seal loses hermeticity silently.
- Use conical (Belleville) washers where the assembly will thermal cycle. Mica does not conform, so any torque lost to cycling turns straight into interface thermal resistance.
Failure modes worth designing against
AN1040 opens with a TO-66 case specifically: a socket-mounted TO-213AA part replaced by a TO-220, leads bent to reach the socket, fastened with a sheet metal screw through a 1/4 in hole containing a fiber sleeve. Bending cracks the package or the bond wires, the oversized hole and the resulting bow shrink the contact area, and the junction runs far hotter than the arithmetic predicted. It is the single most common way a TO-66 sustainment job goes wrong.
The quieter failures are thermal. An insulated joint assembled dry costs 1.4 °C/W. A joint that loses torque over a few hundred hours costs more. Silicone oil migrating out of older greases deposits on cooler surfaces elsewhere in the assembly, which matters if there are relay contacts or optical parts nearby. And because the flange is at collector potential, a punctured insulator does not degrade performance, it shorts the collector to chassis.
What you can still buy, and what it costs
This is where the TO-66 package decision usually gets made. Two manufacturers still list the outline in distribution: Microchip, through the former Microsemi high-reliability line, and Central Semiconductor. Prices and availability below were checked at Mouser on 8 August 2026.
| Part | Manufacturer | Ratings | In stock | Price, qty 1 | If not stocked |
| 2N3585 | Microchip | NPN, 300 V, 2 A | 80 | $28.50 | stocked |
| 2N6317 | Microchip | PNP, 60 V, 7 A, 90 W | 97 | $35.18 | stocked |
| 2N5664 | Microchip | NPN, 200 V, 5 A | 100 | $43.78 | stocked |
| 2N3791 | Microchip | PNP, 60 V, 10 A | 99 | $44.55 | stocked |
| 2N5430 | Microchip | NPN, 100 V, 7 A, 40 W | 119 | $49.29 | stocked |
| 2N3584 | Microchip | NPN, 250 V, 2 A | 0 | $37.63 | 60 pcs due 7 June 2027 |
| 2N6295 | Microchip | NPN, 80 V, 8 A, 50 W | 0 | $28.40 | 100 pcs due 2 Aug 2027 |
| 2N3739 | Microchip | NPN, 300 V, 1 A, 20 W | 0 | $41.58 | 50 week lead time |
| 2N6298 PBFREE | Central | NPN, 60 V, 8 A, 75 W | 0 | $12.20 at 90 pcs | 2 week lead time |
| 2N6297 TIN/LEAD | Central | PNP, 80 V, 5 A, 50 W | 0 | $15.07 at 60 pcs | 6 week lead time |
| 2N4233A PBFREE | Central | NPN, 80 V, 3 A, 75 W | 0 | $10.09 at 120 pcs | 5 week lead time |
Table 4. TO-66 availability snapshot, Mouser, 8 August 2026. Sixty-nine TO-66-2 bipolar line items were listed in total.
Three details in that table change how you plan a build. Central minimum order quantities land on multiples of 30 because the TO-66 case ships in antistatic sleeves of 30 devices; the packing quantity, not the price break, is what sets your minimum buy. Lead times on non-stocked Microchip parts run 48 to 50 weeks, and two lines show incoming stock dated mid-2027, so a shortfall discovered at build time is a schedule problem measured in quarters rather than weeks. And the price delta against a modern plastic equivalent is roughly two orders of magnitude.
Lifecycle status is worse than the stock figures suggest. Central Semiconductor has issued three product discontinuation notices covering most of its TO-66 catalog: PDN01224 (2N3054A, 2N6315, 2N6316), PDN01247 (2N3583, 2N3584, 2N3585, 2N3738, 2N3740, 2N3741, 2N6317, 2N6318), and PDN01250 (2N3054, 2N3055, 2N3766, 2N3767 and several hundred others, on a stock-only basis). Check the notices before you design a Central part number into anything with a ten year service life.
Compliance is the flip side of that scarcity. The Microchip high-reliability TO-66 parts are listed as non-RoHS, which is exactly what they should be: terminals are solder-dipped Sn63/Pb37 over nickel-plated Alloy 52, and the die attach uses high-lead solder. RoHS-compliant matte tin versions exist, ordered with the (e3) suffix at Microsemi or the PBFREE suffix at Central. Central also sells the same die tin-lead plated under a TIN/LEAD suffix. Getting the suffix wrong is the difference between a part that passes your RoHS declaration and one that does not.
Qualification is the reason the price is what it is. The stocked parts are qualified to MIL-PRF-19500 slash sheets at JAN, JANTX and JANTXV levels: /384 for the 2N3584 and 2N3585, /402 for the 2N3739, /455 for the 2N5664, /379 for the 2N3791, /461 for the 2N6211 family. No plastic package carries that qualification. If the requirement traces to a JAN part number, no amount of thermal argument gets you out of the metal can.
Should TO-66 stay in your design?
Work down this list and stop at the first line that matches.
- The build must trace to a JAN, JANTX or JANTXV part number, or the application demands a hermetic cavity: keep TO-66 and buy the Microchip part. Budget $28 to $50 per device and 48 to 50 weeks if it is not on the shelf.
- You are sustaining hardware whose chassis already carries the diamond footprint and the socket: keep TO-66 and buy the Central equivalent if it is still active. Verify against the PDN notices first.
- You need more than about 40 W in a high-voltage linear stage, above 250 V, and second-source stock matters: TO-66 is still competitive, mainly because qualified inventory exists.
- None of the above, and dissipation is under 50 W: move to TO-220 or TO-247. One screw instead of two, an isolated variant available off the shelf, and a bill of materials cost measured in cents.
- You are retrofitting a modern part into an old TO-66 socket: use an adapter bracket that presents a flat mounting surface, and route the leads with flexible wire. Bending leads to fit the old socket is the documented way to crack the die.
Frequently asked questions
Is TO-66 the same as TO-213AA?
Yes. JEDEC re-registered TO-66 as TO-213AA when it consolidated the flange-mounted headers with 0.200 in pin spacing into the TO-213 family in Publication 95, September 1976. TO-123 became TO-213AB and TO-124 became TO-213AC. Datasheets from Microsemi and Central still print both names on the same drawing.
Which TO-66 pin is the base?
On a standard two-lead TO-66 bipolar transistor the two pins are base and emitter, and the metal case is the collector. Microsemi labels them T1 base, T2 emitter, case collector. Three-lead and four-lead variants of the same footprint exist, so check the drawing rather than assuming from the outline.
What screw size does a TO-66 package use?
Two 4-40 screws. The Aavid Thermalloy TO-66 mounting kit No. 4813 specifies 4-40 UNC-2A by 1/2 in pan head screws with shoulder washers, number 6 flat and lock washers, and 4-40 nuts. onsemi characterizes the flange interface at 6 in-lb of mounting torque.
How much power can a TO-66 transistor dissipate?
Between 20 W and 90 W at a case temperature of 25 °C, set by the die rather than the outline. The 2N3739 is rated 20 W, the 2N6317 and 2N6318 reach 90 W. In free air with no heatsink the number collapses: the 2N3584 and 2N3585 are rated 2.5 W at 25 °C ambient.
Can I replace a TO-66 with a TO-220?
Electrically, often yes. Mechanically it needs an adapter. The footprints share nothing: TO-66 uses two mounting holes about 23 mm apart and 0.200 in pin spacing, TO-220 uses a single hole and 0.1 in pins. Bending TO-220 leads to reach a TO-66 socket is the failure case onsemi opens AN1040 with. Use a bracket and flexible wire instead.
Are TO-66 parts still in production?
Narrowly. Microchip still builds the ex-Microsemi high-reliability line, and Central Semiconductor lists a handful of active parts on 2 to 6 week lead times. Most of the rest is discontinued, including everything on Central notices PDN01224, PDN01247 and PDN01250. Around 69 TO-66 bipolar line items remained listed at Mouser in August 2026.
The decision
Specify the TO-66 package when hermeticity or MIL-PRF-19500 qualification is a requirement you cannot argue away, and when you can absorb roughly $30 to $50 per device and a lead time measured in quarters. In that case, budget the joint properly: 0.9 °C/W through greased 2 mil mica at 6 in-lb, 4-40 hardware, conical washers, and a junction temperature target near 110 °C rather than the 200 °C on the datasheet.
Everywhere else, design it out. Below 50 W in a commercial product, a TO-220 or TO-247 part gives comparable thermal performance for a hundredth of the cost, with one screw and an isolated option on the shelf. Run the arithmetic in the worked example above with your own numbers. If the package contributes half the thermal path and the heatsink contributes the other half, as it did at 6 W, the choice of package is not what is limiting the design, and a cheaper package will do the same job.
