New Mars Forums
You are not logged in.
- Topics: Active | Unanswered
Announcement
#1 2026-10-31 21:30:06
- kbd512
- Administrator
- Registered: 2015-01-02
- Posts: 8,573
Fluidics-Based Computing
I've asked myself what kind of computer can withstand intense radiation in the Van Allen Belts or in deep space for many years, and many tens of thousands of thermal cycles, long enough to travel to another star. The answer is that vanishingly few solutions exist if the timetable is measured in decades or centuries. However, one kind of compute core is remarkably durable over these time spans.
To be clear, fluidics are not like-for-like replacements for modern electronics in terms of achievable clock speeds or data throughput rates. Realistic clock speeds range from single-digit Hertz to 5 kiloHertz. They won't hold a candle to any smartphone in terms of computing power, even the original models from 20 years ago. However, most navigation, life support, and vehicle control functions require tiny amounts of processing power and bandwidth, which is why the Orion space capsule uses chips that were "state-of-the-art" about 20 years ago. It doesn't need or meaningfully benefit from more power.
The Apollo Guidance Computer (AGC), an early mostly solid-state electronic device from the 1960s, featured clock speeds near 2 MegaHertz, which sounds blazing fast compared to 1-5kHz fluidics-based devices, except that the systems and peripherals it communicated with were all ponderously slower than it was, and largely still are. Moreover, its serial bus communication would not be any faster in practice than a modern parallelized-logic fluidics-based computer operating at 1-5kHz.
A fluidics computer could replicate the AGC's logic using a machined block of many layers of diffusion-bonded Silicon (high-purity Silicon, but nowhere near as pure as a microchip or photovoltaic cell, therefore less costly) about the size of a playing card deck. It would be powered by Supercritical Nitrogen at 350bar and a tiny pump delivering acoustic pressure waves. Logic gate feature sizes would be 50-100nm, though thankfully RAM and ROM "bits" can be much smaller than that. The device would be just as fast as the AGC in practice through parallelization, but would use only 1 to 2 Watts of pumping power compared to the AGC's 55 Watts, and it'd weigh 1lb or less compared to the AGC's 70lbs, but likely 15lbs for a complete AGC as we know it with robust protection, mounting hardware, and the user interfaces to enter and execute commands. It would still function at temperatures and radiation flux values that would completely destroy the AGC, not over a year, but over a century or several centuries.
Using Orbital Angular Momentum (OAM) and 5kHz clock speeds for monitoring critical life support functions, such as detecting pressure hull breaches (compartment pressure drops or detection of an interruption to the flow of sN2 through fluid channels carved into the hull) for sending signals back to a central compute core aboard a large interstellar spaceship, uncompressed plain digital data could be sent from per-compartment compute nodes at a rate of 32,000 bits per second. Latency would be limited to sN2's speed of sound at 350bar, about 510m/s, but that is fast enough for the in-compartment compute node or for a response from centralized voting computers to seal off any affected compartment(s).
We've been discussing fusion reactors and fusion rocket engines lately, and one interesting conclusion I've come to is that fluidics-based devices operating at 5kHz, physically mounted to the walls of a Tokamak, may actually be several times faster than a network of fiber optics and a supercomputer located hundreds of meters away to avoid the intense radiation fields created by the superconducting electromagnets, for controlling plasma instability, simply due to network latency and all the overhead of modern digital electronic computers with operating systems. It may actually be fast enough to respond to micro-tears in the plasma because it's integrated into and powered by the walls of the tokamak. These devices would still need to communicate with each other using fiber optics vs pressure waves, but close proximity to the reactor, enabled by the ability to withstand temperatures and magnetic fields that would instantly destroy any electronics-based control system, makes a world of difference.
So... as unimpressive as their low clock speeds and data throughput rates are, not dying because a power cable was cut, network latency killed throughput, or a battery backup failed is a lot more impressive.
Repost of an old Scientific American article on the topic:
Scientific American, December 1964, pages 80-88
From the article:
How reliable are fluid amplifiers compared with their electronic counterparts? A number of comparative tests have been made under various conditions. In the range of what may be called ordinary temperatures -- between 150 degrees and minus 50 degrees centigrade -- electronic amplifiers give a more reliable performance. But at higher or lower temperatures the fluid device is more reliable; for example, at 250 degrees C. or minus 150 degrees C. Its reliability is five times greater than that of an electronic device. When they are compared in various environments of ionizing radiation, the story is much the same: the electronic systems are more reliable at ordinary radiation levels but the fluid systems surpass them under exposure to high radiation levels, such as in or near a nuclear reactor. The fluid systems also stand up much better under vibration -- an important factor in a rocket, where the guidance system is subjected to the severe vibration accompanying the rocket's takeoff and acceleration. Strong vibration tends to weaken the solder joints and other connections' of an electrical system; it has little or no effect on a fluid ampliHer. One test model has come through 3,500 million cycles of high-speed vibration at 260 cycles per second without any detectable impairment of its performance.
Fluidics and Electronic Circuit Logic Equivalents
FLUID CIRCUIT performs the operation of dividing by 10 in an all-fluid digital computer: for every 10 input pulses circuit delivers one output pulse. Input pulses enter from above the plane of circuit through circular, bumplike hole attached to straight channel running from top to bottom just to right of center. The 10 identical logic elements, or modules, are arranged in a series of five pairs, three at left and two at right. Each pair contains two steady input streams (small sausage shapes), two outputs (small circular shapes attached to short straight channels), eight control jets (small teardrop shapes) and eight open vents (large circular shapes). Shallow channels were chemically etched in Dycril plastic by a special process, called Optiform, developed by Bowles Engineering Corporation. For this photograph the channels were filled with white paint to increase the contrast. Normally the working fluid is air.
Another small tech write-up from "obsolete.studio" an organization specializing in the "intersection between art and science":
Form for Fluid Computer
Univac FLODAC explanation PDF:
FLODAC-A PURE FLUID DIGITAL COMPUTER
A very nice write-up showing more logic gates and how they're used:
Fluidic Computation Kit: Towards Electronic-free Shape-changing Interfaces
A collection of PDF presentations from 2007:
Coding and Computation in Microfluidics
If you've worked on cars with automatic transmissions, you may be familiar with the complex valve body that wholly or partially governs the transmission's operation. This is a very primitive fluidics-based computing application that uses hydraulic pressure from the transmission fluid to manipulate the path the fluid takes through the valve body. It looks a lot like a circuit board, except carved into metal, because that's exactly what it is.
Offline
Like button can go here
#2 2026-10-31 21:50:31
- tahanson43206
- Moderator
- Registered: 2018-04-27
- Posts: 26,353
Re: Fluidics-Based Computing
This post is reserved for an index to posts that may be contributed by NewMars members.
This new topic appears to me to have significant upside potential, and I'm not talking about some distant future. Vehicles being designed by GW Johnson right now would seem to me to be strong candidates for computing hardware as robust as the opening post suggests they might be. Johnson's designs are intended to last for many years in space transit operations. It seems to me they won't require ultra-high-speed computers to manage the onboard systems. Modern communications systems operate at high data rates, but a vehicle that takes Earth years to complete an operations cycle (such as some of GW's space tugs would do) may not require such high data rates to maintain contact with it's owners.
What GW's designs most certainly ** will ** require is computing hardware that can be counted upon to remain functional despite the great variety of assaults the Universe we live in will throw at it over decades or longer.
Index:
(th)
Offline
Like button can go here
#3 2026-10-31 22:01:45
- kbd512
- Administrator
- Registered: 2015-01-02
- Posts: 8,573
Re: Fluidics-Based Computing
To the best of my knowledge, the only major modern-day application of fluidics is "computer-on-a-chip" (a literal piece of plastic that a human or automated machine injects fluids to test for a very specific interaction) biomedical lab testing, where it is used extensively. I guess hydraulics and transmissions are also major applications, but nowhere near as sophisticated as biomedical fluidics. Regardless, think of this as SCADA using supercritical fluids instead of electronics. Most of the same basic concepts used in electronics still apply to fluidics, but you're using an actual fluid instead of a stream of electrons. If you're more worried about end-of-the-world reliability than data throughput rates for systems that don't require high data rates, then I think there's something here worth pursuing.
Offline
Like button can go here
#4 2026-11-02 08:49:46
- NewMarsMember
- Member
- Registered: 2019-02-17
- Posts: 2,149
Re: Fluidics-Based Computing
This topic has potential to provide a great deal of useful information. In order for that to happen, something like a lesson plan would be helpful. GW Johnson posted about the automobile transmission as an example of a single purpose logic device.
The "computer" in the transmission is an analog computer. Its flow path diagram with the losses-or-not is the analog to a circuit diagram with resistors. It can only respond to the inputs it was configured-for. It is not a general computer.
I do not know, but I suppose that a grid of either-this way or-that-way fluidic elements could represent binary 1's and 0's. If so, then a fluidic digital computer is possible. Compared to electronics, it would be physically quite large. The energy to drive the fluid pressures is applied to the fluid outside the fluidic elements themselves, so unlike electronics, the waste heat should be outside the computing elements themselves. Such a thing should be a general computer, like an electronic digital computer. How it would be programmed is an utter mystery to me.
GW
(th)
Last edited by NewMarsMember (2026-11-02 08:50:32)
Recruiting High Value members for NewMars.com/forums, in association with the Mars Society
Offline
Like button can go here
#5 2026-11-03 21:24:08
- kbd512
- Administrator
- Registered: 2015-01-02
- Posts: 8,573
Re: Fluidics-Based Computing
GW,
If you're wondering how a fluidics-based computer could store a program as "ROM" (Read-Only Memory), it's rather simple:
A pin/bump/divot etched into the metal substrate directs the flow of gas in one direction and represents a "1", whereas its absence represents a "0". A system clock, or "timer circuit", is simply a flow-path feature that induces an oscillation or pressure wave in the gas at a set frequency. This is no different from a series of transistors redirecting electron flow in an electronic circuit. We use piezo crystals that vibrate at a specific frequency and transmit electrical vs pressure wave pulses for system clocks/timer circuits. If you look at the "AND" logic gate in the image I posted, it works no differently than the electronic equivalent on the same row.
Whereas electronic circuits must follow the rules of electron movement through a conductive pathway, fluidics-based logic gates/circuitry must follow the rules of fluid dynamics for the selected fluid.
The specific substrate I've selected for my fluidics-based computing concept is Silicon. It's high-purity Silicon, but nowhere near as pure as the Silicon required by a microchip or photovoltaic cell, so far less energy is required for the macro-scale fluidics "chip" / compute core. The selected fluid to carry signals is supercritical Nitrogen. The physical wiring connecting this computer to the associated sensor network is sealed Nitinol wire. Since the Silicon core and external Nitinol wires are both hermetically sealed and pressurized to about 350bar, metal diaphragms carry pressure waves to and from the core.
Thin wafers of Silicon are cut from the ingot, same as for microchips or photovoltaics, and alternating layers are cut or etched to create flow/logic channels inside them. Every other layer is a solid Silicon wafer. After the "logic stack" is cut into the Silicon, the individual Silicon layers are diffusion-bonded back together to form a complete compute core. This device is then encased in hard rubber or plastic and placed inside a metal box that protects the compute core from accidental impact, no different from the giant Aluminum box that encased the Apollo Guidance Computer (AGC). The only difference is the box size. A fluidics-based computer is nowhere near as large as the AGC for the same logic gates because all the surrounding electronics-based components are absent in a fluidics-based computer. If you're wondering why we'd use high-purity Silicon, it minimizes acoustic reflection, and thus "cross-talk", at least for the clock cycle range and feature sizes I'm interested in.
A thermoconvective pump driven by a radioisotope like Carbon-14 or Nickel-63, combined with a nozzle-type diffuser (to ensure cyclic one-way flow through the circuitry), can deliver 1kHz (50-100 micron flow channels) to 5kHz (15-20 micron flow channels) pressure waves through the supercritical fluid without any electrical power using very minor temperature deltas (supercritical fluids exhibit large changes in volume from minor temperature fluctuations). Using Nickel-63 would involve embedding 2-micron-thick foils in power-loop channels to maximize thermal power transfer into the sN2. Heat pipe systems can also deliver external power without radioactive substances present inside the compute cores. If the thermal power input is waste heat from life support systems aboard the starship, then there is no added waste heat generation. However, this requires a connection to a heat/pumping source that is "always on", like a pure thermal fission or fusion reactor. This thermal power source's distribution lines or "power cables" would be embedded in the ship's hull plating, meaning waste heat is siphoned from a power reactor and/or life support systems (such as the chemical bioreactors used to synthesize sugars/proteins/fats for the crew), distributed through channels embedded in the hull structure, and therefore does not require significant additional tubing/wiring mass.
If the heat source is embedded in the circuitry, the circuitry must be damaged or destroyed to affect the heat source. If putting very thin radioisotope strips into the circuitry is too hazardous because a compute core blown open by a MMOD strike could aerosolize a beta emitter, then we stick with heat pipes for thermo-acoustic power delivery from centralized nuclear power sources. Strictly speaking, nuclear heat is not required. Almost any decent heat source will do. Simple waste heat from the life support systems is entirely sufficient to run a highly sophisticated distributed fluidics-based compute architecture that handles all life support control and shipboard system control functions. It's entirely possible to transmit data between the distributed compute nodes at speeds up to 32,000 bits per second. Sealed Nitinol wiring will connect sensors to compute cores within a compartment. To minimize through-bulkhead penetrations, pressure transducers/signal repeaters can re-transmit or broadcast data through bulkheads.
A secondary compute architecture using pure photonics will archive data from the fluidics-based compute cores by writing recorded temperature/pressure/atmospheric composition/life support equipment health data to 5D data crystals. These data crystals are like extreme-density ROM chips. A crystal the size of the US Dime can hold many terabytes of data. Photonics-based compute devices use a million to ten million times less power than electronics for equivalent throughput, but macro-scale life support systems require pumping power and heat. Fiberoptic cables don't like being flexed back and forth, and unlike a fluid channel carved into a metal plate, they require multiple materials and devices to communicate data over a network. They're delicate in a way a hole in a metal plate isn't. The dopants mixed into the fiber, and the cladding surrounding the fiber, are the most difficult things to replicate aboard a starship. There's a world of difference between fabricating extreme-purity doped glass or plastic fiber (which will also degrade long before the mission is over) and carving a channel through a wafer of Silicon, or forming a new Nitinol tube and filling it with Nitrogen. A pure photonics-based computer chip or display/monitor with a crystalline substrate display is effectively a different technology and won't meaningfully degrade from GCR exposure or crack from CTE mismatches between the variety of metals used in electronic computers. Photonics-based computers would need to use radioisotopes like Carbon-14, which can convert electron emissions into light using phosphors or other materials. Compartment lighting will also use Carbon-14 or Nickel-63 sources, so no wiring, LEDs, or switches are needed to generate white light. The casings for these light sources can be much more robust than the casings for fluidics-based compute cores, so accidental exposure to an aerosolized betavoltaic source is far less likely.
Anyway... For moving valves or monitoring pressure and temperature sensors to control macro-scale, thermal-power-driven devices, fluidics-based computers that consume waste heat offer more and better options for amplifying pressure to drive actuators, despite seemingly ponderously slow clock speeds. Efficiency and speed are relative. If all your power comes from waste heat you intended to dump overboard anyway, does it matter if a computer, unaffected by GCR, electrical conductivity, or thermal degradation, is slower and uses more power than electronics that would need to be completely remanufactured many times to sustain a mission to another planet or star? I think ultimate durability and longevity, with little to no maintenance, matter a lot more. Beyond that, 100% of additional electrical power consumed by an electronics-based compute infrastructure is ultimately converted into waste heat.
You had to first generate electricity, an inherently less efficient process for a thermal power source like a reactor core, because heat is driving a turbo-generator and efficiency peaks near 50% for that conversion process using rotating turbomachinery at temperature deltas that high-spec superalloys can survive for 100,000hrs or so. MHD generators do meaningfully better, so 65% thermal-to-electric conversion efficiency. Electrostatics providing 80-90% efficiency demands contradictory temperature conditions exist within the same compact space for fission or fusion reactors (to prevent the device from melting or acting like a giant resistor). I suppose it's doable, but not without cryocoolers, which are very inefficient. Nobody's made it work in a practical way, even though that doesn't rule out future success.
The point is that if your systems are driven purely by thermal and pumping power, you don't suffer an immediate, massive up-front conversion loss. If an electronic computer to control the life support systems only needs 500W of power, whereas a fluidics-based computer requires 1,000W of power, you still generate 1,000W of thermal power to get 500W of electrical power, except it's usually far worse than that for most power reactors, meaning 25-35% efficient. That 500W of power for the control systems was in addition to whatever electrical power your CO2 scrubbers and water purifiers require, and their electric pumps also convert electrical power into waste heat. Imagine that your water and air still require 500W of pumping power, but you only pay that unavoidable thermal power conversion and dissipation penalty once instead of twice. So what if the fluidics-based computer requires twice as much waste heat from your life support system as an electronic system would consume? You're already generating that much waste heat, and you're going to dump it overboard using giant radiator panels. Wouldn't it be better spent doing something useful (powering a computer immune to GCR, CTE mismatch cracking, and thermal degradation from waste heat)?
Ironically, an AGC-equivalent computer based on fluidics is about the size of a deck of playing cards and needs only 1-2 Watts of thermal-pumping power. Despite its much lower clock speed, the fluidics-based computer responds faster (using parallel communication) because the serial-bus communication to the connected electronics-based components was/is even slower than the pressure waves coursing through the sN2. The real-life AGC used 55W of electrical power from an onboard fuel cell, and it weighed 70lbs. Would a modern all-solid-state electronic computer beat the pants off the fluidics-based AGC-equivalent in terms of processing speed and data throughput? It absolutely would, every day of the week, and it still wouldn't matter in the slightest, because the very nature of the hardware the AGC controlled, which includes modern equivalents, wouldn't allow for meaningfully faster overall response times.
Offline
Like button can go here