All about the IR3109 chip

IR3109 Chips

Introduction One of the key components in vintage Roland synthesizers is the IR3109 four stage OTA chip. It enabled polyphonic analog synthesizers to be created due its smaller footprint and it helped create the sound of the 1980’s Roland analog synth, until digital filters took over in 1987. This post explores the technology inside the chip and where it was implemented from 1979 – 1986.

The Technology The IR3109 is intended as a synthesizer filter chip with an exponential CV converter, 4x OTAs, and 4x buffers (one after each OTA). It is very similar to the CEM3320 and SSM2040 chip in objectives, but with slightly different technical detail.  Roland provided a minimal datasheet in the Jupiter-4 service manual: The IR3109 contains four variable transconductance amplifiers designed for VCF applications in musical instruments. The device is equipped with four high input impedance buffers , and anti-log circuitry (VIN to VOUT) which controls the conductance of for amps.

• wide transconductance variable range (1uυ -10mυ)
• low offset voltage (less than +/-3mV)
• high impedance MOS P channel buffer

It is an open design which means the stages can be used independently, so the chip can be configured as LP, HP, All Pass, as a SVF or used as four VCA’s but always under a common cutoff frequency CV which uses the exponential converter. At Roland the chip was used as a 2 and 4 pole low pass filter, 2 pole SVF with HP and LP outputs, and as a dual and quad VCA. At Boss the chip was used in phase shift mode in the PH-2 pedal and the RPH-10, and the IR3109 and D80017A appeared in the the GR-300 and GR-700 guitar synthesizers.

The technical design is assumed to be based on OTA’s that are BA662 cores, and the exponential converter is based around a NPN/PNP pair. The P-MOS buffer transistors are likely to be DMOS devices, which is unusual and may explain the difference in sound to a CEM3320. A Modwiggler thread about the IR3109 engineering is here.  The chip is frequently used with external temperature compensation.

Which synthesizers used it?  Roland started using the IR3109 in the Jupiter 4 from March 1979 to replace the hand matched BA662’s. The implementation was a standard OTA based 24dB filter with no Q compensation, heard in many Roland mono and modular synthesizers during the late 1970’s. The Jupiter 4 filter may sound a bit noisy because the signal levels into the IR3109 were rather lower than the headroom available, this “mistake” was not repeated in later designs.

The development of the Jupiter 8 in 1980 took full advantage of the new IR3109, with both 2 and 4 pole low pass modes, and Q compensation, where the input signal is boosted back into the chip when high resonance levels are dialed in. This compensates for a signal level reduction as resonance is increased.

This was unique at the time (January 1981), and not seen in the contemporary Prophet 5 and OB-X, which used the CEM3320 as per the datasheet. The Jupiter 8 was also different in that the filters would not self oscillate, although this can be dialed in using PCB trimmers. Later on Curtis made filter chips with inbuilt Q compensation; the CEM3372 in 1982 (Prophet 600) and the CEM3328 in 1984.

The IR3109 was then used in the Juno 6 and 60 (1982), the Jupiter 6 (1983) and the early MKS-80 models (1984). The Jupiter 8 configuration was simplified for the Juno’s, with only one mode (24dB low pass) and a reduction in components. However it added the ability to self-oscillate, which was common across all 3 Juno’s, and it retained Q compensation. Whilst the single DCO needed a chorus to fatten the sound, the 24dB filter delivered a fantastic sound.

IR3109 in SH-101

1982 also saw the IR3109 used in a mono-synth, the massive selling SH-101 (over 10,000 sold), for the first and only time. This is a similar configuration to the Juno with a 24dB low pass filter with self oscillation but no Q compensation. Once again it defined the Roland sound of the early 1980’s with some special magic that is hard to replicate without the IR3109.

In 1983 the IR3109 implementation in the Jupiter 6 broke new ground for a Roland filter, and hailed a different approach that would last for another 3 years. It was configured as two 12dB SVF’s with HP and LP modes, lots of good sounding resonance but no self oscillation. A BP mode could be selected by pressing both front panel buttons together. The Jupiter 6 implementation sounds significantly different to the Jupiter 8, especially at high resonance levels, and has both found lovers and haters of the SVF sound.

The resonance of the filter can be adjusted by a PCB trimmer but this is aimed at getting consistent sounds across the 6 voices and its not a way to dial in lots of self oscillation in. Q compensation is included on the first SVF, with the input signal feed into the resonance control OTA, this actually increases the signal level at low resonance settings – the reverse of the technique with a standard low pass filter. There are documented hacks to increase the signal level at higher resonance levels (by feeding the resonance CV into the VCA) and for removing the low resonance Q compensation.

IR3109 inside JX-3P

Roland also released the low cost JX-3P in 1983 at £1000 with a simplified filter implementation based on the IR3109 in 24dB low pass mode. There is no Q compensation or self oscillation, and the resonance is controlled by two transistors rather than a VCA. Full resonance is set quite low in volume, which is noticeable when compared with the Juno range. It can be sorted by adjusting the resonance PCB trimmers to provide a higher signal level from the filter (800mV+ rather than 600mV). The JX-3P has become regarded as having a thin tone due to the mild maximum resonance.

The MKS-80 of 1984 was launched with the same Jupiter 6 circuit but with only one mode (24dB low pass), which was rather expensive and component intensive for just the one mode! So Roland switched over to the IR3R05 chip half way through the production run (REV 5 version), to reduce costs and save PCB space. However the MKS-80 did introduce Roland customers to the sound of a 24dB low pass filter created from SVF’s, which became the basis of their polysynths for the next 3 years. It may also explain why the MKS-80 never sounds quite like a Jupiter 8.

Roland released the 16-voice MKS-10 analog piano module in 1984, using 16x IR3109 in 24dB low pass filter mode with variable cutoff and a fixed low resonance. This 2U rack unit has inevitably become a source of IR3109 chips to rescue vintage Roland synths, like the Jupiter 8. The MKS-10 was quickly replaced by the far better sounding digital MKS-20.

Summary The sound of the early 1980’s Roland synthesizer was dominated by the IR3109 chip, implemented as a 24dB low pass filter, with Q compensation in the Jupiter 8 and Juno polysynths. Each synthesizer has a different implementation and sound, although there are strong similarities between the Jupiter 8 and Juno 6/60.

The final days of the IR3109 in analog synthesizers were in the last SH-101’s manufactured in mid 1987. However the chip lived on inside the Boss PH-2 phaser pedal, which was manufactured until 2000 possibly longer. This was a fortunate for people needing IR3109’s for repairs, as the PH-2 could be bought at low prices in 2010 and the IR3109’s salvaged. The prototype AM8060 filter was built from a PH-2.

With over 50,000 chips manufactured in THD and SMD (possibly more), this was a Roland success story both in terms of sound but also in keeping costs down. The need for further cost reduction was to influence the next stage of Roland filter chip design.

IR3109 SMD chip

The IR3109 in SMD The D80017A chip appeared in the Juno 106, MKS-30 (1984) and the MKS-7 (1985). It was a further miniaturization of the VCF and VCA sections of a synthesizer with a SMD IR3109 chip and two SMD BA662F chips, soldered onto a ceramic carrier dipped in black resin. Whilst the SMD chips have proved long lived, the resin and printed circuit traces were a fatal mix and caused many failures over the years.

Replicas of the D80017A soon appeared from 2000 onwards to keep Juno-106’s alive, typically using LM13700 OTA’s and TL072 JFET buffer op amps. The schematic of the D80017A is the same as in the Juno 6 and 60 (24dB low pass filter), with Q compensation and the ability to self oscillate.

IR3R05 Chip

The IR3R05 In 1984 the IR3109 chip gave way to a re-modelled version of the chip as the IR3R05. This contains two state variable low pass filters, which are configured in series, with onboard resonance control and a VCA. This filter type was first seen in the Jupiter 6, where the IR3109 was used to create two SVF’s, with a second IR3109 performing resonance control.

The IR3R05 packed even more analog circuitry into a 14-pin DIL chip, with the resonance CV control and a VCA with both linear and exponential control. This chip enabled the next range of Roland analog polysynths (JX-8P, JX-10, MKS-70, Alpha Junos, MKS-50) to be manufactured at lower price points and created the Roland analog sound of the mid 1980’s.

However the downside to this implementation was the lack of self oscillation, which reduced the range of sounds that could be created from the filter. This was done because it is hard to tame the SVF’s when they go into self oscillation, with distortion and clipping a real issue. There is no resonance PCB trimmer, which makes it easier to build the synth, as the chips were consistent in quality across voices, but you can not dial it back in without making resistor changes. The IR3R05 can be driven into self oscillation with a larger resonance CV signal, which is what we have done with the AM8105 Super-JX module.

AM8109 JP-8 Filter

AMSynths Modules Back in 2010 we launched two module that replicated the filters in the Jupiter 8 and Jupiter 6, as the AM8109 and AM8060 respectively, using original IR3109 chips. We also located a stock of NOS IR3R05 chips and launched the AM8105, which like the AM8109 had the ability to go into self-oscillation at high resonance settings. The eventual shortage of these chips meant the modules went out of manufacturing, and new additions like the AM8106 did not get released (Juno 106 filter).

In 2022 the availability of a low cost and great sounding IR3109 clone from Alfa Rpar (AS3109) meant we could restart production of the AM8109 and AM8060 modules, as well as new modules such as the AM8101 (SH-101 filter), AM8160 (Juno filters) and AM8180 (MKS-80 REV4 filters). The upcoming AS662 will also enable more Roland VCF’s and VCA’s to be added to our range of modules.

Filter Capacitors The sound of a filter is partially reliant on the type of capacitor used in each filter stage, and how closely matched they are. Expensive polystyrene capacitors sound the best when closely matched at 1%, which means the filter points are the same in each stage, which results in nicer sounding resonance and self oscillation onset. Roland only used these type of capacitors in the flagship SH-7 monosynth and went with cheaper ceramic disc capacitors in nearly every IR3109 implementation.

The Juno-106 uses SMD ceramic capacitors, which may explain a difference in sound to the Juno-60 which uses THD versions. In 2022 high quality PPS SMD capacitors are available for synth designers or THD MLCC ceramic and polypropylene. The AMSynths modules use 2.5% polypropylene capacitors in the AM8109 and AM8060, and 10% MLCC ceramic capacitors in the AM8101 and AM8160 and AM8180.

What do they sound like? There are plenty of demos of Roland analog synthesizer on You Tube, so you can listen to the differences there. Whilst I have owned the Roland Jupiter 8, Jupiter 6, MKS-70, MKS-80, Juno-60 and JX-8P, I never recorded any filter comparison demos! I will upload demos of the AMSynths filter modules onto the You Tube channel.