MEMS Version of Intel 4004 Chip Could be Made to Prove Nanomechanical Computer Designs

Ralph Merkle, Robert Freitas and collaborators have shown that using only links and rotary joints a Turing-complete computational system can be created. A version of this design could be built using existing and popular 2 micron MEMS technology to make a chip that would equal the 4-bit Intel 4004 CPU. The Intel 4004 came out in 1971 and was the first commercially available microprocessor by Intel.

Universal combinatorial logic has been demonstrated with the design of a NAND gate, while sequential logic, mimicking electronic flip-flops and sufficient to create memory has been demonstrated using cells combined into shift registers.

Above – Part of a molecular mechanical logic gate. This molecular machine would have 120695 atoms, 87595 carbon and 33100 hydrogen, and occupies a volume of about 27 nm × 32 nm × 7 nm.

The design approach is far simpler than any other mechanical Turing-complete design. The design avoids almost all sliding friction so the design has the potential to be more power efficient than any previous design. In fact, simulations suggest that molecular-scale implementations of the described system would be far more power efficient than conventional electronic computers.

Silicon-based electronic computers are everywhere and are cheap and powerful so why would we want nanomechanical computers? Many research groups are currently investigating mechanical, electromechanical, and biochemical alternatives to conventional semiconductor computer architectures because of their unique potential
advantages. Mechanical systems can withstand much higher temperature and radiation exposure than their electronic counterparts, and hence may be useful in certain niche applications.

MEMS Version is Possible

Flexure joints provide an alternative implementation with similar general performance to pivots. Flexures have the advantage that in many cases, particularly with MEMS, they are easier to fabricate and often more reliable than fully functional pivot joints.

A systematic method of design allows all necessary locks, balances, bell cranks, support links, and transmission links to be implemented in only two layers of material.

What could be possible with conventional MEMS technology?

The minimum feature size of the popular Multi-User MEMS Processes (MUMPs) commercial program is two microns. If the flexures are two microns in width then a pair of transistors would cover an area of 640 × 1070 microns. A silicon die 2.8 cm square could contain the mechanical equivalent to 2,200 transistors, which is the transistor count in the 4-bit Intel 4004 CPU.

Nanomechanical Version Would Be 100 Billion Times More Energy Efficient Than Todays Computers

A nano-mechanical computer could be 100 billions of times more energy efficient than current computers. A previous version of the nanomechanical design calculated potential efficiency. The nanomechanical computer is well-suited for implementing physically reversible logic gates. Reversible logic gates are one alternative technology that can, in principle, sidestep fundamental limitations of complementary metal-oxide-semiconductor (CMOS) transistors, and thus facilitate computers that operate with vastly reduced energy dissipation.

They remove the need for gears, clutches, switches, springs and make the design easier to build. Existing designs for mechanical computing can be vastly improved upon in terms of the number of parts required to implement a complete computational system. Only two types of parts are required: Links, and rotary joints. Links are simply stiff, beam-like structures. Rotary joints are joints that allow rotational movement in a single plane.

Simple logic and conditional routing can be accomplished using only links and rotary joints, which are solidly connected at all times. No gears, clutches, switches, springs, or any other mechanisms are required. An actual system does not require linear slides. Any traditional 2-input logic gate, including AND, NAND, NOR, NOT, OR, XNOR and XOR, can be created directly from the appropriate combination of locks and balances. Reversible gates can also be created using only links and rotary joints, and a Fredkin gate is also demonstrated.

Universal combinatorial logic and sequential logic together are known to be enough to make a general-purpose (Turing-complete) computational system. Subject to practical limits of time and memory (as in any computer), such a system can compute anything that can be computed.

Reversible Computing

A mechanical computer designed as described herein has the potential to provide 1 trillion GFLOPS/Watt, over 100 billion times more efficient than conventional “green” supercomputers, which currently provide about 18 GFLOPS/Watt.

The Landauer Limit of 4 x 10^-21 J per logical operation is quite large compared to the energy lost to friction via a NAND gate of the link- and rotary-joint style (3.9 x 10^-26 J per logical operation). It makes little sense to have such an efficient mechanism, but to then operate it in such a manner that power losses due to bit erasure reduce its effective efficiency by several orders of magnitude.

Reversible computing is a well-known way to circumvent the Landauer Limit. Using reversible computing schemes, bits are stored, even if technically no longer needed to produce the final output, so that the penalty associated with bit erasure is not incurred. It would be trivial to modify the described NAND gate to be reversible.

Arxiv – Mechanical Computing Systems Using Only Links and Rotary Joints

SOURCES – Arxiv, Ralph Merkle, Robert Freitas
Written By Brian Wang.

14 thoughts on “MEMS Version of Intel 4004 Chip Could be Made to Prove Nanomechanical Computer Designs”

  1. K, so… for the slightly computer-science minded types, a few thoughts. 

    [1] SPEED

    Computing is quoted in lots-of-MHz. The newest i–9 can chug to 5,700 MHz unofficially, liquid cooling, 8 cores, 16 threads, gazingabytes of memory, Optane pseudo-memory to terabytes, and so forth.  

    Intel, AMD, everyone design around individual gates that “do their thing” a few picoseconds a pop. Might even be sub-picosecond, I’m not keeping up with it.

    Compare that to a nanomechanical thing. Sliders, stirrups, pivots, motion. Accelerations, inertia, momentum, torque.

    Seems that even if shrunk down to the lowest possible nanometer feature-size, ought to be slow. 


    Analogous to electronic circuits, a “gate” needs:

    • inputs
    • output(s)
    • a state map
    • power

    Even if I’m playing around with rubber bands and tinker-toys (which technically can model anythig that this article purports!!!), the combination of inputs (providing power sort-of) and “the power supply” seems to be minimally necessary for real-world gate operation. So, how’s MEMS do it?

    [3] FAN OUT

    Turns out a “big deal” when designing anything logic-oriented is to deal with the limits of what a gate can provide as output power. E.g. if all gate inputs are a load of ‘1’ unit, and the simplest gates can output ‘5’ units of power, well … technically they might reasonably drive up to 5 strung-together downwind inputs.  But that’s all. 

    Just saying,
    GoatGuy ✓

  2. That is one weird idea. I like it! There are already proposals for computing with phonons (lattice vibrations, not a typo for “photon”).

  3. You are correct, but to be clear the 97% savings is over the Landauer Limit, not over the energy current processors use. Current processors use far, far more power to switch a gate than the Landauer Limit requires. Reversible computing offers (as a ballpark figure) 97% savings against the theoretical minimum. Since current processors are nowhere near that limit, the savings as compared to a conventional processor is 99.9%+.

    Also, note that the Landauer Limit is temperature-dependent. It is defined as ln2*KT, where T is temperature. So, by running the system colder, you can get essentially arbitrarily more savings — it just depends on how cold you are willing to keep it. Run it at liquid nitrogen temperatures and you would get an additional savings of several-fold over room temperature. Run it at liquid helium temperatures and you would get an additional savings of more like 100X over room temperature. Of course, there is a cost to the refrigeration, but assuming it still results in a large net savings, something like 99.99%+ is probably possible (and adding a few more 9’s there might be reasonable — I’d have to check on the ratio of ln2*KT to what modern processors use).

  4. This 100 billion number is pure BS. Reversible computation is only free if there is zero I/O or unlimited memory.

    Reversible Computation isn’t Free

    It’s like Amdahl’s Law. You can get rid of 97% of your compute system via reversible computing, but you’re still with that irreducible 3%. So you really reduced it by 33x.

  5. But there are some interesting potential applications for explosives that compute, and translate the product of the computation into a shaped shockwave. Imagine, for instance, a hand grenade that does explosive image analsysis, with the outcome of the computation being hypersonic metal jets in the direction of identified foes only…

  6. In one sense it might be possible. If the storage of that “information” builds a regular crystal then the later corruption of the crystal should be able to generate energy. If that is reversible then you have a storage system. But it is not exactly information as we know it, it is just order.

  7. Remember that mechanical nano-computers were intended more as an existence proof than a serious proposal. Because molecular transistors and the like were viewed as too hard to analyze to persuade the skeptical.

    It would be ironic if they turned out to be worth building anyway.

  8. intel 4004 has 2,300 transistors, intel i9 has 7 billion transistors. Your layering will need to be 3 million deep.

  9. 100 billion times more efficient? So I’m thinking we can now layer up said Intel 4004 level logic to produce a real 3D block of computronium.

    1. Massively parallel multi-core designs with a level of inter-core communication impractical with 2D layouts
    2. A way to get the same total number of components as the much smaller, but heat-limited-to-2D electronics
  10. “The physics are wrong” … because ultimately all stored energy that is “mechanical” is either compression (or tension), or mass-in-motion. Springs and their analogues, and moving masses (centrifuges, tops, bullets).  Or gravitational: rocks on a mountain, helium balloons, undersea gas bags.  

    MEMS devices have no special, new, undiscovered “corner” on compressional or centripetal energy storage. At least not compared to springs or fibers under tension, and rotary centrifugal devices.  

    Just saying,
    GoatGuy ✓

  11. Seeing this, I immediately thought the impossible – storing energy in the form mechanical energy, and releasing it by the means of reversible computing. Basically a new kind of battery. The physics are wrong, but I don’t know how.

  12. one cool thing about mechanical computers; no leakage. so 35% power savings at least for the same operation (assuming they switch at the same “reverse bias” voltage). ENIAC 2020, here we come!

    I’ve seen similar designs to leverage this effect fabbed in diamond before, a sort of microscopic diamond based vacuum tube. interesting ideas all of them, especially as they are radiation immune so space based applications are on the table in a big way.

Comments are closed.