AI summary by TwinMind
Why IBM's Superconductor Computer Failed
Asianometry
21m 19s
Historical Context and Josephson Effect Discovery (1962)
- 1962: 22-year-old Cambridge grad student Brian Josephson predicted mind-blowing superconductor phenomena
- Josephson studying under Phil Anderson (Nobel Prize winner 1977) at Cambridge
- Anderson teaching broken symmetry theory and its connection to superconductivity
- Josephson discovered that Cooper pairs (not just single electrons) can tunnel through insulating barriers
- Josephson effect: supercurrent flows through barrier without resistance below threshold current
- 1963: experiments verified Josephson's predictions despite initial skepticism from John Bardeen
- 1973: Josephson won Nobel Prize at age 33 alongside Ivar Giaever and Leo Esaki
Early Superconductor Computing (1950s-1960s)
- 1950s: concepts for superconductor computers existed using vacuum tubes
- 1955: Dudley Buck at MIT proposed cryotron - superconducting switches
- Simple design: tantalum wire with copper wire wound around it
- Operated in liquid helium to achieve superconductivity
- Switched off by applying current to copper wire, creating magnetic field
- 1959: Buck died unexpectedly at age 32 from viral pneumonia/edema
- Industry progress continued:
- IBM created faster thin film version
- GE put 2,000 superconductor switches on single integrated circuit
- Silicon planar integrated circuits offered superior performance without liquid helium refrigeration
IBM Josephson Junction Development (1960s-1970s)
- Post-1963: IBM began working on thin film devices for digital computers
- Jurgen Matisoo at IBM named devices "Josephson Junctions" to differentiate from failed cryotron
- Early devices: two pieces of lead or niobium separated by thin insulating oxide barrier
- Memory circuit design: current circulating forever in superconducting loop
- Direction of current flow indicates 1 or 0
- Different from DRAM which stores charge in capacitor
- Late 1960s: IBM initiated full research program
- 1970s: IBM focused on integrating multiple Josephson junctions onto single chip
- NSA contributed funding hoping system could break Soviet codes
Manufacturing Challenges
- Fabrication used same tools as silicon ICs (lithography, deposition)
- Major challenge: uniform insulating barrier 1-4 nanometers thick
- Silicon dioxide advantage: easy to grow on silicon substrate
- Josephson materials required sputtering deposition method in low pressure oxygen
- Manufacturing variations created reliability issues
- Josephson junctions gave weak uncertain outputs vs. silicon's robust outputs
Performance Promises vs. Reality
- Switching speed claims: as fast as 10 picoseconds
- Power consumption: orders of magnitude less than silicon (several microwatts)
- 1977 predictions: 10,000x performance improvement, $30 megabit storage chip by 1985
- Reality by 1980: 50-100 chips per core immersed in liquid helium at 4 Kelvin
Silicon Competition and Dennard Scaling
- Mid-1970s: Robert Dennard proposed scaling theory (Dennard's Law)
- Smaller transistors required proportionally smaller voltages
- Power consumption per unit area remained constant despite higher transistor density
- PC Revolution fueled rapid microprocessor innovation
- Silicon became formidable competitor without liquid helium requirement
Technical Limitations Discovered
- Asymmetric switching speeds:
- Resistive to superconducting: <10 picoseconds (good)
- Superconducting to resistive: few hundred picoseconds (limiting)
- Practical circuit speed gated to ~1 GHz
- Early 1980s: silicon chips approaching/surpassing these speeds
- "Punch through" issue: nonzero probability of gate failing to reset to zero voltage
Memory Problem
- IBM never achieved large enough working memory
- Josephson memory cells required inductors and resistors, making them huge vs. DRAM
- DRAM cell: just capacitor + transistor
- 1983: IBM achieved only 1 kilobyte memory
- 2006: NEC Japan achieved only 4 kilobytes (20+ years later)
Project Termination
- 1980: IBM council of experts evaluated project progress
- 12 out of 15 members voted not to continue
- 1983: IBM cancelled project after 15 years and $100 million spent
- ~150 person team in Yorktown Heights, NY dispersed to other IBM projects
- Japan's NTT picked up technology but ended project after 8 years
Current Applications and Future
- Today: few significant Josephson devices in use
- SQUIDs (Superconducting Quantum Interference Devices):
- Measure weak magnetic fields
- Used in neuroscience, metrology, submarine detection
- Mostly analog systems with low device counts
- New hope: Rapid Single Flux Quantum (RSFQ) logic
- Invented in Moscow mid-1980s
- Uses quantized magnetic flux pulses instead of voltage states
- Current dominant form of superconductor digital logic
