By " traditional parametric engines", I mean the ones used by Digikey, Mouser, alltransistors.com, or even the NTE Source Research site (SRI).

AI can be ChatGPT or the simple Google Gemini AI results that are generated in a trad. Google search. I don't subscribe or pay for any AI service so I am limited free, basic AI services. If the subscription services are more accurate, let us know!

Anyway, I work on vintage audio gear, and many devices like transistors are no longer manufactured (or reliable, cost-effective new-old stock sources are rare). Even sub-sub-substitutes are now "Obsolete" at DK/Mouser! So one must come close somehow with what is still available.

In another Reddit thread, I noted some issues with a 1980 Hitachi HA4700 amp that has some damage to the pwr amp board components.

https://www.reddit.com/r/audiorepair/comments/1ursnq7
I am having a tough time with biasing and thermal runaway in the channel in which I had to replace damaged transistors with substitutes. I do have transistor testing meters that provide some useful info.

I thought part of the issue with biasing is the use of NTE parts. On forums like AudioKarma, NTE has subpar reputation.

For example, in looking for replacement 2SB648, I first used NTE SRI, and that spits back NTE2520. (I sometimes use NTE parts simply because a local store still has new-old stock NTE parts in stock). So I soldered in two new NTE2520 for the two blown 2SB648s. While this is progress, I still have biasing and thermal runaway issues.

In ChatGPT, I entered "2SB648, best substitute transistor". And it spat back "2SA1220A" (choice no. 3) as the best still-available alternative. I do have new NEC 2SA1220A's in my parts bin so I swapped out NTE2520s for the 2SA1220A. Still had the same thermal/biasing issue, but in addition, one of the resistors began to smoke and another T0-126 began to get very hot (thermal camera help). I had the unit on a varaic so not much damage. But the AI-suggested 2SA1220A don't seem up to the task.

What's odd is ChatGPT does not suggest the NTE2520 as an alternative even though it suggests other obsolete/discontinued devices.

So out of curiosity, I entered into ChatGPT "NTE2520 vs 2SB648". And it spat back:

"The NTE2520 is not equivalent to the 2SB648, but it is one of the better replacement options if an original 2SB648 isn't available."

I looked at the original (blown) 2SB648 device, the NTE2520 and the NEC 2SA1220A. The 2SA1220A physically looks a little less "beefy" than than the orig. Hitachi 2SB648 or NTE2520. More importantly, the hFE parameter between the NTE2520 NEC 2SA1220A are quite different. See photos.

I have not re-installed the NTE2520's back in to confirm that the 2SA1220A's are, indeed, the cause of new problems. (May follow up if I do that experiment).

Back to trad. parametric search for substitutions ... not sure it is any better/worse than AI.

Trad. parametric search engines (alltransistor, SRI for NTE, or DK/Mouser ) do bias towards (rank better) non-obsolete parts. And AI, depending on how you word the query, will rank with obsolete parts in context (inclusive).

 

LLMs are only going to spit back at you what they can gather from their training, which may include comments scraped from the general internet. Assume any answers they give you are wrong, and its up to you for confirmation (by testing or through traditional sources).

I ran your query, "2SB648, best substitute transistor", on Grok. Here's the result:

The 2SB648 (and variants like 2SB648A) is a silicon PNP epitaxial transistor in a TO-126 package, primarily used as a low-frequency, high-voltage amplifier.

Key Specifications (from datasheets)

• Polarity: PNP
• Vceo: -120 V
• Vcbo: -160 V (or up to -180 V in some listings)
• Ic (continuous): -50 mA
• Power dissipation (Pc): 1 W
• hFE (DC current gain): 60–320 (binned as B/C/D groups, e.g., 100–200 for C)
• ft (transition frequency): ~140 MHz
• Complementary NPN: Typically 2SD668 / 2SD668A
• Package: TO-126 (similar pinout to many medium-power drivers)

It's an older Hitachi (now Renesas) part that's somewhat obsolete and can be hard to source in small quantities from major distributors, though limited new/old stock appears on eBay, AliExpress, or specialty sites.

Best Substitutes

The strongest modern/recommended equivalents (prioritizing close voltage, current, gain, frequency, and package match) include:

• KSA1220 / KSA1220A (onsemi/Fairchild) — Excellent match; widely recommended. TO-126, -120V / -160V versions, good for audio/driver stages. Complementary is KSC2690 / KSC2690A.
• 2SA1220 / 2SA1220A — Very similar Japanese original equivalent.
• Other good options: 2SA1380, 2SA1406/1407, 2SB649, 2SA1021, or 2SB631K (check hFE and exact voltages for your circuit).

Always verify the full datasheet for your specific application (e.g., audio amp driver, preamp, etc.) and test in-circuit if possible. Pay attention to:

• hFE bin (match gain group if critical).
• Heat sinking (TO-126 needs proper mounting).
• Pinout (usually E-B-C, but confirm orientation).

For availability, search major distributors like Digi-Key/Mouser for KSA1220A (often in stock) or the originals on secondary markets. If you're repairing something specific (e.g., vintage audio gear), provide more circuit context for finer recommendations.

 

I ignore whether this vintage audio repair is a hobby or an income generating activity, but you might consider investing in a transistor curve tracer.
The problem with these small transistor testers is that, by its very nature, can only be tested at both low voltages and currents. Exactly the opposite operating region of a class AB power amplifier.
Thus the values you obtain may bear little resemblance with those in actual operation.

 

This may possibly be way off base, but it might even be a connection issue. I have come across devices with a different connection arrangement. I do not know if that applies in this instance, but certainly it is worth being sure that it is not a connection issue.

 

This may possibly be way off base, but it might even be a connection issue. I have come across devices with a different connection arrangement. I do not know if that applies in this instance, but certainly it is worth being sure that it is not a connection issue.

I don't think that this is off base. At the very least, you need to check which lead is BASE using a DMM.
Then you need to confirm which is COLLECTOR and EMITTER.

 

I did have a solder bridge in the with two pins (base to collector??) , Q702, and that fixed this problem I noted earlier:

"...but in addition, one of the resistors began to smoke and another [component] ,TO-126, began to get very hot (thermal camera helps). I had the unit on a varaic so not much damage. "

So, after fixing the bridge, I left the both 2SA1220A's in. And I can, once again, run the amp, at full power, with the two R-ch bias fusible R's to pwr trans (R738R for Q714R; and R739R for Q713R) disconnected. This means the pwr trans don't turn on. And I can TRY to do some tracing, in the driver section, as to what is turning them on full-blast.
I still have the original biasing / thermal runaway issue on the R-ch. (The L-ch is fine).
"Connection issue" might be the elephant in the room. It may be that the 46-year-old amp is losing trace (PCB trace) integrity. This means even gently probing voltages, with amp on full power, can cause blown parts or fuse R's to smoke.

Recall that I purchased the amp with quite a bit of damage to the R-ch. But L-ch was fine, as was the preamp, PSU/reg, LEDs and lamps. All the mains 800mA fuses were good and the unit powered on but no protection relay click. Replacing many out-of-spec metal fuse R's in the L and R channel does now allow relay to engage, and I can fine tune L-ch biasing current and DC offset very nicely. L-ch behaves itself.
But even after all that, the amp will destroy itself if it not balanced. So, low-resistance (68-100 ohm) fuse resistors, in the pwr amp section. nor the PSU main 800mA fuses, won't protect your scarce and $$ and beefy Sanken pwr trans. substitutes (now even those are unobtanium). Troubleshooting the R-ch, I blew both NPN and PNP Sankens, and I was being careful with a variac!