The ultimate TB-303 transistor sound comparison


The characteristic sound of the TB-303 has been subject of discussions for years. There have been several discussions especially about the transistors used and their impact on the sound of the TB-303. The original TB-303 was manufactured approx. from Q4 1981 until Q3 1983. During this period of time Roland changed the board design as well as some of the components for various reasons. In total Roland used 4 different types of transistors during production time. This article analyses the timeline of TB-303 and its revisions. Additionally it shows what types of transistors were used during production and how the sound differs from a non technical perspective. Hi-Res-Sound samples are included in the documents section so you can make up your own mind what could be the ideal transistor for a TB-303 or a clone. Because the listening experience is subjective, the conclusion reflects my personal opinion and is not scientific at all. I leave the judgment to you.


Level of difficulty:


The author of this tutorial states that you are exposed to low voltage. Although we take great care about our tutorials neither the author nor obsoledo takes any responsibility for insuries, false information or any damage caused by this tutorial. If you have any doubts please contact a professional service. Please be careful.






Serial Number:




Date of Production:



No accessory included or considered.


Step 1

First a little bit of background information on this project. I worked on several TB-303s and clones throughout the last 20 years. At some point it became obvious that some of them sounded better than others. This started my interest to investigate why this is. After several clone projects with the latest being the RE-303 replica and endless discussion on the internet some of the most important parts that influence the typical 303 sound were identified. Just to name a few the 2SK30-AO is important for the hollow square-wave, the sumida coil used in the PSU creates an unstable power management that causes voltage drops influencing the sound, the Roland custom IC BA662 adds an extra level of dirt to the VCA and the usage of film capacitors is necessary if you like the filter to be smooth just like in the original 303. It also has become a common understanding that the PNP-tranistors (2SA733) play an important role in the filter circuit.

Step 2

Revision analysis:
Over the years I collected information about some TB-303s that passed my bench and I noted their serial numbers and the components used inside. Although all of the previously mentioned components responsible for the sound stayed the same, the NPN-transistors changed. In total 4 types of transistors could be identified. According to my list, Roland started the production of the 303 by using the 2SC1685 transistor, then used the 2SC536 for most of the production time. I suppose that during the production at some points they went short on 2SC536 and used 2SC1815 instead because the usage of these can not be followed over along long period of time. There were also units with 2SC536 and 2SC1815 used in the filter cascade only. The 2SC945 according to my records is suspected to be the least used one. Overall is seems that the average TB-303 is equipped with the 2SC536.
So what is so different about the other transistors and why were they used?

Step 3

Test setup:
To spot the differences in sound a TB-303 replica was used with soldered sockets for all NPN-transistors (Q34, Q24, Q25, Q29, Q30, Q11, Q13, Q14, Q15, Q16, Q17, Q18, Q19, Q20, Q23, Q35, Q37, Q40, Q41, Q32, Q1, Q2, Q3, Q4, Q7). This excludes Q42 in the PSU and Q204 on the sequencer board. The replica, a RE-303 is an exact built of the TB-303 using all original components and PCB layout. For each transistor a recording session was taken. For that a pattern with typical 303 characteristics was created containing slides, accents and tied notes. Using sockets made it easy to exchange the transistors for each recording session. A very important aspect was to recalibrate the filter according to Roland specs after each exchange. This ensured that the sound was not influenced by the VCF trimmer but only by the main potentiometers. Additionally to the 4 types of transistors used by Roland I added a very common 3904 to see if there is a difference in sound.

Step 4

In total 6 recordings were created for each transistor in 24-bit 96kHz. These were then converted to lossless FLAC. The following potentiometer settings were used.

Recording 1: Tune=5; Cut Off=0-10-0; Resonance=0; Env Mod=0; Decay=0; Accent=1
Recording 2: Tune=5; Cut Off=0-10-0; Resonance=5; Env Mod=5; Decay=5; Accent=1
Recording 3: Tune=5; Cut Off=0-10-0; Resonance=10; Env Mod=0; Decay=0; Accent=1
Recording 4: Tune=5; Cut Off=0-10-0; Resonance=10; Env Mod=10; Decay=10; Accent=1
Recording 5: Tune=5; Cut Off=0-10-0; Resonance=10; Env Mod=5; Decay=5; Accent=1
Recording 6: Tune=5; Cut Off=0-10-0; Resonance=10; Env Mod=10; Decay=0; Accent=1

The settings of the potentiometer are also indicated in the audio files you can find in the documents sections or at the soundcloud link (e.g. T5_C0-10_R0_E0_D0_A1.flac = Tune 5; Cut Off = 0-10-0; Resonance=0 etc.)

Step 5

The result of course is as mentioned earlier subjective. From my point of view the differences in sound are almost not noticeable. The only marginal difference I could spot was that the 2SC1815 creates a smoother filter than the other transistors. This might also explain why Roland used them in some revisions in the filter section only. From other gear built at the same time as the 303 I know that many Japanese manufacturers preferred the 1815 for audio application. The Toshiba data-sheet supports this claim by mentioning the purpose of the transistor to be for audio applications. One might investigate the differences on a more technical level but I believe this is not necessary because I feel confident that it does not matter at all what transistors you use for the NPN ones.

Step 6

This is why I also added the general purpose 3904 transistor to see if this makes a difference. If you listen to it one might think "Wow, yes this is totally different.". Actually what makes it different was that by using this type of transistor it was simply not possible to do the filter calibration using the VCF trimmer because the range was not sufficient. This means that the filter is not opened enough to match the other recordings. I did not alter the circuit to fit the others but I am positive that one might even use these transistors instead of the Japanese types to achieve similar results in sound as with the other transistors.

Documents & Files





No comments

You must be logged in to post a comment.


View More Results…

Category Browser

Open all | Close all

Filtered Search