Why

The real world is analog. The mathematics of quantum electrodynamics explain the nano-scale world with extreme precision. The mathematics can predict the probabilities of future events, thus we know the mathematics is sound. Time, space, and probabilities of state-transitions are all complex number continuous variables (at least as far as we’ve been able to see).

The momentum ($p = \hbar k$) and Energy ($E = \hbar \omega$) are sometime said to be quantized, but since $k$ and $\omega$ can take on any value for a free particle, I don’t think the quantization is general. For bound fermions, however, the energies and momentum are quantized. Only fixed values of $k$ and $\omega$ are possible. Two electrons (a fermion, or spin $1/2$ particle) cannot exist in the same space and the same energy. That’s why when you bring two atoms close together, like a crystal, the wave-functions interact, and the discrete energy level split into bands of allowed energy.

Since the real world is analog, there will always, to the end of human time, be a need for engineers that can make analog circuits. Independent of technology.

Abstraction principle

In software there is an Abstraction Principle for how software engineers should think about software development. The general principle is Don’t repeat yourself. We violate that often in analog circuit design.

70 years ago the first transistor was made, but it was only with the work of Mead and Conway in the 1980’s and the pure foundries, like TSMC, we got EDA companies. Today Cadence (1988), Mentor Graphics (1981) and Synopsys (1986) dominate the EDA industries.

For the past 40 years the way we make analog circuits have not changed much. Every 5 ish years we have to change technology (180 nm to 55 nm to 22 nm). Every time we change technologies we have to repeat our self. We have to create the schematics, and the layout all over again. For every technology we have to spend months to years to again make analog circuits that work. There is virtually no reuse of analog circuit design between technologies.

Analog circuit design have a intimate, entangled, unbreakable bond with the integrated circuit technology. A current proportional to temperature (PTAT) circuit made in 55 nm CMOS from TSMC is unlikely to work in 55 nm CMOS from UMC. All technologies require tweaks to the analog circuit.

We rely on a process design kit for a technology with simulation models that describe as detailed as possible the physical behavior of a transistor. Since no two factories, or companies, have exactly the same manufacturing process, two transistors from different companies will not match in their physical behavior.

For digital circuits, we’ve abstracted away the transistor. We know we can configure two PMOS and two NMOS into a 2-input NAND. A NAND is a circuit where when both inputs are high, the output is low. Otherwise the output is high. Based on a NAND we can construct any digital circuit. To transition to a new technology it’s just one cell we need to make. In practice we make a few more “standard cells”, but that’s due to area, speed or power optimization.

Analog circuits play between high and low. The currents, charges and voltages between nodes, and how they change, create insanely complex interactions. To my knowledge, no-one has found a general abstraction level for analog circuits.

Circuit abstraction layers

Although there is no general abstraction level for transistors, we can reuse ideas between technologies. A PTAT current may have almost same schematic and the same interface in two technologies, however, the size of transistors, resistors, bipolars and capacitors, and their physical properties will be different.

I think it’s pointless to find a general abstraction level, for devices (transistors, resistors, capacitors, inductors, bipolars, diodes). Those will always be different, they are in the real world, and their physical behavior is tied to how the integrated circuit is manufactured.

I think we should focus on reusing the idea. We must disentangle the analog circuit idea from the technology.

I’ve been working on the problem of “idea reuse” since 2009, and have made some progress. Below is a status, and my current thoughts. I think for each circuit abstraction layer the choice of data source for the design and layout should change.

Devices should definetly be compiled, probably the same for cells (level shifters, standard cells).

For cells it does depend on the complexity. A cell that is too complex to be easily understood by looking at the netlist, should not be considered a Cell, but rather, a Block.

A typical Block can be an OTA, comparator, bias current generator, voltage regulator or similar. At some point a Block consists of only Blocks, with few analog interfaces between them. As soon as analog interfaces are few, I think it makes more sense to transition to SystermVerilog.

Status Abstraction Design Layout Why
:construction: Chip SystemVerilog digital Complex connections, few analog interfaces
:construction: Module SystemVerilog digital Large amount of digital signals, few analog signals
:warning: Block Schematic programmatic Large amount of critical analog interfaces, few digital
:white_check_mark: Cell Netlist/JSON compiled Few analog interfaces, few digital interfaces
:white_check_mark: Device JSON compiled Polygon pushing
:white_check_mark: Technology JSON/Rules compiled Custom for each technology

See My thoughts on analog for more information on Technology, Device and Cell abstraction layers.

Block

An analog block, like a PTAT current source have large amount of critical analog interfaces, and a few digital signals.

From books, papers, or in rare instances a bright mind, we get the “idea” for the PTAT current source.

A schematic, like the one below, is how we store the idea. Analog designers are trained schematic readers and most can immediately understand how the circuit works.

Making the idea work in a particular technology, however, is fundamentally a different task. With a quick glance an analog designer can see that the circuit won’t work in 55 nm for core transistors if the supply is low (1.0 V).

The VD1 is always a diode voltage above ground (0.9 V ish at low temperature). $V(VBN,VD1)$ will at least be a threshold voltage (0.5 V ish). $V(VBNC,VBN)$ will be at least 0.2 V, and $V(AVDD_CV,VBNC)$ will be 0.2 V ish. Thus the minimum AVDD_CV of the circuit is $ 0.9 + 0.5 + 0.2*2 = 1.8 V$. As a result, in 55 nm with core transistor the “idea” is a “bad idea”, but in the same technology with IO transistors (3.0 V) it may be a “good idea” and could be worth a try.

To answer whether the “idea” is good in a technology, the analog designer must tweak all the width, lengths, multipliers, of each device and run through hundreds of simulations. Afterwards, the schematic may look the same, but all devices will have tweaked values.

Porting to another technology will again change device values, but the schematic will look the same.

I think it would be a good idea to separate the “idea” from the technology implementation. Each technology port is about 5 years from each-other. Accordingly, it’s rare that the port is done by the same person as last time. That means, a new person will try and copy the old idea, and maybe even redraw schematic again.

Humans are error prone, and it’s easy to draw a wire wrong, or insert a wrong component, or forget a power down transistor. Had the “idea” been stored in a technology independent format, there would be less chance of introducing human errors.

Idea

What I’d love to do is to use Xschem schematics to store the “idea”, then store the technology specific items, like exact transistor types, widths, lengths, in another format (YAML or python).

The “idea” schematic and technology layer could go through a compiler to generate a technology specific schematic.

Exactly how to do that, I’m not sure yet, but I think that’s my next step.