------------[ The bucket brigade device: An analog shift register ]-------------
Publication date: 2024-11-14
Last updated:     2026-04-26
 

With computers, delaying audio is easy, but that only works if you have
megabytes of RAM and a good ADC/DAC... but this wasn't always possible.


These are the TCA350, a fully analog delay line: 

[IMG /misc/bbd/device.jpg ]

[IMG tca.png ]

The two clock signals consist of non-overlapping negative pulses, each making
every second MOSFET conduct. Additionally, both clock lines are coupled
to every other capacitor, which helps pull charge from one stage to the
next. 

A sample is only taken in during the T1 phase, it's really the charge difference
between two capacitors that moves through the device. Because of this, the
185 transistors only hold 92 samples: 

not a lot, but it's enough to add an audio echo or reverb effect. 

The chip doesn't need too many external components: 

[IMG sch.png ]

[IMG complex.png ]

[IMG beat.png ]

The yellow trace is the input, and blue is the output 

There's a lot of really nasty sampler kickout, that would need to be carefully
filtered out in a real device: 

[IMG close.png ]

Looking at the schematic, there's a capacitor directly connected between
a clock line and the output transistor's gate... so this is to be expected.


I wanted to see how the device was built, so I cut open one of the cans:


[IMG metal_boxes.jpg ]

Click for a clean, hi-res version 

This photo mostly shows the aluminum layer, which is used to connect all
the transistors. 

The squares marked with yellow are connected to pins of the chip. The lower
left ground pin is connected to the bulk silicon, ensuring that the intrinsic
"body" diodes of the transistors never conduct because --- in typical 1970s
fashion --- the chip uses voltages below ground, not above it. 

There is a single MOSFET transistor in the red box, with the gate coming
in from the left, and the drain and source going off the chip. This is T187
on the schematic, and is responsible for amplifying the output right before
it leaves the chip. 

The actual delay line (center) is almost completely covered in metal. 

To expose the silicon, I scratched up the chip to remove it's passivization
layer, and etched the metal with phosphoric acid: 

[IMG silicon.jpg ]

During this I accidentally broke part of the chip, so it's not a perfect
view. 

These are the two photos aligned on top of each other: 

[IMG both.jpg ]

Blue: metal, yellow: silicon 

Most of the chip is n-type silicon, some regions have been converted to
p-type, which is visible in the photo as the yellow squares. The junctions
between the two types act as diodes, which are always be reverse biased
and don't conduct. 

With a few exceptions, the silicon is covered in silicon dioxide (glass)
which insulates it from the metal. In these places, p-type wells act as
a separate conducting layer. 

[IMG bbd.png ]

... but that doesn't mean they don't affect each other: 

When the metal runs over p-type regions, the oxide forms a capacitor. 

Things get interesting when metal runs over the gap between p-type regions.
If the metal is sufficiently negative, it can pull positive carriers into
the n-type gap forming a temporary channel between the p-type wells. 

In other words, it's an (p-channel) MOSFET transistor. 

(If you need a refresher, lcamtuf has a really good explanation of semiconductors)


[IMG components.jpg ]

Today, these devices are completely obsolete --- but a spin-off, the CCD
camera sensor, is still used today. Instead of transistors and capacitors
like the TCA350, they directly move charge trapped in inversion "bubble"
using three sets of MOSFET-style gates. 

This requires an rather inconvenient overlapping 3-phase clock, but works
amazingly well: a good CCD can move charge between a million pixels without
loosing a single electron.