Asianometry Podcast Summary

Episode: Ternary Computing: Theoretically Better than Digital

Host: Jon Y
Date: December 18, 2025

Episode Overview

In this highly-requested episode, Jon Y dives into the theory and history of ternary (base-3) computing, contrasting it with binary systems. He explores why ternary computing has long tantalized theorists and researchers, the rare historical attempts to realize it, modern technological twists, and why—despite its mathematical allure—it still hasn't overtaken binary's dominance. Major themes include the technical advantages and challenges of using three-state "trits" instead of two-state "bits", historic Soviet experimentation, ongoing research efforts, and why ternary logic could be relevant again in the age of AI.

Key Discussion Points and Insights

What is Ternary Computing, and Why Trits?

• Introduction to Trits vs Bits [00:02]
  • Ternary digits ("trits") use three states, e.g., -1, 0, and 1 (balanced) or 0, 1, 2 (unbalanced), instead of binary’s two.
  • Higher bases allow more information per digit—so theoretically, data encoded in trits can be denser and hardware simpler than with bits.
  • "A 10 digit number encoded in binary requires 40 bits, but only 21 trits."
• Mathematical Efficiency
  • The optimal theoretical base for representing information is Euler’s number (~2.718), but base 3 (ternary, efficiency: 0.366) is nearly optimal and better than binary (0.347).
• Additional Theoretical Perks
  • Native support for negative numbers, simpler comparison logic, and, as Jon jokes, aesthetic appeal:
    • "Minus 1, 0, 1, I know Thanos would approve." [01:40]

Historical Attempts: The Soviet Setun

• Early Innovations [03:27]
  • The first known ternary calculator dates to Thomas Fowler, 1840, later reconstructed by historians.
  • In the 1950s-60s, the Soviet Union’s "Setun" computer at Moscow State University became the only production ternary computer.
• Setun’s Origin Story [06:41]
  • A grudge prevented MSU from getting a standard binary machine, so the lab was ordered to build their own—leading to ternary experimentation.
• Ternary Ferrite Core Memory
  • Inspired by the magnetic core logic of the time (LEM1), Setun modified the architecture to let two ferrite cores store a trit, giving their system a ternary state.
• Political Drama & Demise [11:23]
  • Despite successful demos and brief production, political rivalry and competition from binary projects led to Setun’s shutdown. The later Setun-70 was never widely produced.
    • Notably, designer NP Brusentsov defended ternary's philosophical and technical merits:
      • "He always believed that ternary is superior to binary... binary's all or nothing. Nature deviates from human reasoning and creates unworkable paradoxes in thought." [13:20]

Theoretical vs. Practical Limitations

• Disparity between Theory and Reality [15:40]
  • Setun’s alleged simplicity was not borne out—using two cores to store a single trit isn’t more efficient than storing two bits.
  • "Two cores can also store two binary bits... so one might argue that Setun probably would have been better off as a binary computer."
• Complexity in Circuit Design
  • Doug Jones, University of Iowa: Some ternary adders are actually more complex (by 62%) than binary, even after adjusting for data density.
• Human Factors
  • Ternary logic is harder for developers and users to grasp, as demonstrated by confusion over SQL’s three-valued logic (true, false, null).

Hardware Hurdles

• Device-level Obstacles [19:00]
  • Binary works well with MOS transistors and two voltage thresholds. For ternary, distinguishing three levels in hardware is difficult—requires precise voltage control and higher manufacturing complexity.
• Notable Early Research
  • 1970s-80s: H.T. Mufta’s team in Canada experimented with CMOS ternary logic, crafting several prototypes but hitting barriers like power inefficiency and complexity.
  • "No matter what the circuit outputs, there's a current constantly going, eating power and generating heat." [21:40]
• Modern Device Candidates
  • Memristors: Can, in theory, support multi-level logic, but emulation complexity remains an issue.
  • Carbon Nanotube Transistors: Recent Chinese research demonstrated a device with three distinct current states, showing promise for ternary SRAM and other circuits. [24:51]
    • "The author said that the CNT transistor can be easily manufactured and produce several ternary circuits with it, like an sram."

Ternary’s 21st Century Relevance: AI Inference

• Neural Networks and Quantization [27:30]

  • Modern AI models benefit from reduced precision to save power, memory, and computation time.
  • Aggressively binarized models lose too much accuracy, while ternarization (three-state weights) offers richer representation with minimal compute/memory cost increase.
    • "Ternarization is similar, quantizing everything to trits... there is more accuracy. Moreover, having a zero lets you skip every multiplication action involving that zero."
  • But there are tradeoffs; going "full ternarization" sacrifices too much information. Hybrid schemes are under active research.
    • "Never go full ternarization. The information losses are too drastic. Mapping both 5 and 50 as one doesn't make sense." [29:40, paraphrasing Tropic Thunder]

• Huawei’s New Patent [32:10]

  • Huawei recently patented a ternary logic gate for AI applications—using three groups of transistors with precise voltage thresholds to encode -1, 0, and 1.
  • Manufacturing such gates is complex due to the need for ultra-precise doping and construction.
    • "I do wonder about the complexity of manufacturing three groups of transistors so close together." [33:21]
    • But Jon lauds the ingenuity and sees the patent as a sign of lingering interest in ternary, especially for AI Accelerators.

Notable Quotes & Memorable Moments

• Jon Y, on ternary’s aesthetic appeal:
  "Minus 1, 0, 1, I know Thanos would approve." [01:40]
• Setun designer NP Brusentsov, on ternary logic:
  "It allows for more nuance by moving beyond the simple yes or no. To him, binary's all or nothing. Nature deviates from human reasoning and creates unworkable paradoxes in thought. After all, only a Sith deals in absolutes." [13:20]
• On ternary’s practical difficulties:
  "So one might argue that Setun probably would have been better off as a binary computer." [16:55]
• On AI quantization:
  "Never go full ternarization. The information losses are too drastic. Mapping both 5 and 50 as one doesn't make sense." [29:40]
• On Huawei’s patent:
  "Huawei’s work is a sign that despite binary’s dominance, ternary retains its theoretical allure. Can ternary revolutionize AI inference or compute or whatever? In theory, maybe. But those theories will remain as such until someone goes out and build it." [34:18]

Key Timestamps

• 00:02 — Introduction to ternary logic, trits vs bits, base efficiency math
• 03:27 — The first ternary computing devices, early history
• 06:41 — The Setun story: Soviet politics, technical concepts, ferrite core memory
• 11:23 — Setun’s political sabotage and shutdown
• 13:20 — NP Brusentsov’s philosophical defense of ternary
• 15:40 — Why ternary’s theory hasn’t translated to practice
• 19:00 — Hardware-level obstacles: transistors, device innovation
• 21:40 — Canadian CMOS ternary research and its limits
• 24:51 — Carbon nanotube devices for ternary logic
• 27:30 — Ternary logic and AI model quantization
• 32:10 — Huawei’s ternary logic gate patent, manufacturing questions
• 34:18 — Conclusion: ternary’s theoretical allure vs binary’s practical dominance

Summary

Jon Y delivers a detailed, engaging exploration of ternary computing, weaving together theory, history, and present-day research. While ternary logic theoretically promises denser information, more natural negative numbers, and hardware simplicity, practical obstacles—ranging from device physics to programmer habits—have kept it sidelined. The episode closes noting that, while binary still reigns, advancements in AI and new device research may yet give ternary a new lease on life—“in theory, maybe”—but only “until someone goes out and build it.”