At the bottom of this page is a video showing the key principle in action

What the ADC can do

It’s about power and accuracy. The FFT below shows a third harmonic distortion of -88.5 dB or 14.7 bits.

SI Neural ADC FFT- 1.4MS/s 370uW SFDR 14.7bits

Here is the SiliconIntervention ADC compared to other reported technologies  – the scatter plot below is taken from Boris Murmann’s database of reported ADCs Compared to other 0.15 and 0.18 technology devices the SiliconIntervention ADC is shown in red:

Implemented on an advanced process such as 28nm or 22nm the ADC will have even lower power.

The Principle of Operation

Analog circuits process signals as voltages, currents, charge, pulse density or even time. All those parameters are analog (continuously variable) quantities. One of these analog properties is chosen to represent the signal, and useful operations are performed as accurately as possible on those signals.

Digital circuits process signals as numbers. The electronics is only asked to determine if a signal is “high” or “low”  – if it can accurately do that, then it can process digital data. Just about the only thing we need to worry about in the digital circuit is how fast it can run.

Digital signal processing is a powerful technique partly because it works in the abstract: it does not matter how “high” or how “low” the signal is exactly, that distinction only makes the digital number. The interaction of the digital numbers that are an abstraction “above” the level of the electronics is what matters and where the power of digital signal processing excels.

Is it possible to do analog signal processing in the abstract? That is, not directly proportional to the continuous quantities in the network of wires? The answer is yes it is – and we at SiliconIntervention do that.

One small step to abstract analog is well known: differential signal processing. We do not worry about the absolute voltage (or current, or charge etc) because the signal is a difference of two quantities. The signal is a function not of one, but of two analog quantities. It’s the difference of two quantities and that makes the absolute values not matter. The signal is somewhat abstracted from the parameters of the network.

SiliconIntervention uses a function of four quantities. That means a signal in the SiliconIntervention ADC is expressed over a multiple of four wires not one or two. When we encode an analog quantity over four “conventional signals,” that is over four parameters in the network, we achieve a further degree of abstraction and we can make the signal not only independent of the absolute offset of the network parameters (like differential signals) but also independent of the absolute value of the parameters.

  • The key feature of the encoding method is that the signal value does not settle vs time, it is the signal’s signal-to-noise ratio that experiences the finite settling time. Essentially the probability of the signal being correct increases with time as the neurons (op-amps) settle.
  • This allows the ADC to partially concurrently settle – it does not sequentially settle  – hence it is faster.
2 – Example of Quad Signal encoding

The underlying math of what is happening in the circuit…

There is a precedent for this idea. The bridge network used for decades is a step beyond differential because it is ratiometric: it carries its own reference with it so to speak.

Signals in the SiliconIntervention ADC also carry their own reference with them  – they do not need to know what an absolute voltage (or current or charge…) is.

The innovative step is the mapping of the signal into the 4N parameters that carry an analog real quantity in the network and 8N parameters that carry complex analytic signals. That map creates what is called an invariant subspace and what that means is that any linear operation on the signal preserves the signal, and what that means is that we don’t need to wait for the amplifiers in the circuit to settle in order to continue processing the signal. The settling time of the amplifiers in the system only affects the signal-to-noise of the signal (SNR) and not its value.

  • Settling time of the signal is the mismatch of settling times of the amplifiers used in the circuit. This is what analog designers have been doing for decades: work with the mismatch of elements not their absolute value. We bring that to the time domain;  prior to our innovation, settling time was a first order effect  – we simply had to wait, now we don’t.
  • Settling time continues to affect, to the first order, meaning we have to wait for the SNR to reach the final value, but, waiting for signal-to-noise to settle means we can begin processing the best estimate of the signal in the next stage and there is a degree of concurrent (not purely sequential) operation in the circuits. This is why we call these circuits “Emergence from Noise” circuits.

The video below shows how the signal, imposed upon four wires, is insensitive to the delay of the op-amp.