Read the original article here.

---

Understanding Computer Memory: A Core Component in Building Systems

Computer memory is a fundamental concept in computer architecture. Often synonymous with terms like RAM (Random Access Memory), main memory, or primary storage, it serves as the computer's short-term workspace, holding data and instructions that the processor needs to access quickly for immediate use. Unlike slower, high-capacity mass storage (like hard drives or SSDs), memory prioritizes speed, acting as the CPU's primary interface to the information it requires to execute tasks.

In the context of "The Lost Art of Building a Computer from Scratch," understanding memory goes beyond just knowing its purpose. It involves appreciating its physical implementation, historical evolution, different types, how data is organized within it, and the critical role it plays in system stability and performance.

Core Concepts

At its most basic level, computer memory stores information in the form of binary digits, or bits (0s and 1s).

Bit (Binary Digit)

The smallest unit of data in computing, representing one of two possible states, typically 0 or 1.

These bits are organized and accessed in specific ways:

• Memory Cells: Individual physical locations within the memory chip, each capable of storing one or more bits. Most memory types store one bit per cell.
• Words: Memory cells are typically grouped into larger units called words. The size of a word (word length) varies depending on the computer's architecture (e.g., 8, 16, 32, 64 bits). The CPU processes data in chunks related to its word size.
• Addresses: Each word (or sometimes byte, depending on the architecture) in memory is assigned a unique numerical address. This address allows the CPU or other components to locate and access specific data within the memory.
• Address Space: The total range of unique addresses available in a memory system. If an address is N bits wide, the system can address 2^N distinct locations (words or bytes).

The Evolution of Computer Memory: A Historical Journey

The history of computer memory is a story of the relentless pursuit of faster, denser, and more reliable storage. Understanding this evolution provides valuable context for how modern memory works and the challenges faced by early computer builders.

Early Systems (1940s)

The very first electronic computers had extremely limited memory. The ENIAC, for instance, used thousands of vacuum tubes, each storing a single decimal digit. This was volatile storage suitable for holding intermediate results of calculations, but not for storing programs or large amounts of data.

• ENIAC (1940s): Utilized vacuum tubes to store a small number of numerical values. Highly volatile and physically large.

Acoustic Delay Lines (Early 1940s)

One of the earliest practical forms of computer memory involved delaying sound waves. Developed by J. Presper Eckert, this method used a tube filled with mercury with transducers at each end.

• How it Worked: Electrical pulses representing bits were converted into sound waves by a quartz crystal transducer at one end. These waves traveled through the mercury, reflected off the other end, and were received by another transducer, which converted them back to electrical pulses. These pulses were then re-sent into the tube, creating a circulating loop that "remembered" the data.
• Significance: Provided a way to store more than just a few numbers sequentially.
• Limitations: Limited capacity (a few thousand bits), sequential access (you had to wait for the data to propagate), sensitive to temperature and mechanical vibration.

Tube Memory (Mid-1940s)

Alternatives to delay lines emerged using cathode-ray tubes (CRTs), similar to old television screens.

• Williams Tube (1946): Invented by Fred Williams, this was the first known random-access memory (RAM). It stored bits as electrical charges on the surface of a CRT screen. An electron beam "read" or "wrote" the charges.
• Selectron Tube (1946): Another CRT-based memory developed at RCA.
• Significance: Introduced the concept of random access – the ability to read or write any piece of data directly by specifying its address, without having to go through other data sequentially. This was a major leap.
• Limitations: Volatile (data lost power), relatively low density, expensive, and notoriously sensitive to environmental noise, making it unreliable.

Magnetic Core Memory (Late 1940s - 1970s)

A major breakthrough came with magnetic core memory, which became the dominant form of main memory for over two decades.

• How it Worked: It used small, ferrite (magnetic) rings, or "cores," threaded with wires. Passing current through the wires in specific directions could magnetize a core in one of two directions (clockwise or counter-clockwise), representing a 0 or a 1. Reading involved attempting to flip the core's magnetic state; if it flipped, it generated a pulse on a sense wire, indicating its previous state.
• Significance: Non-volatile. Data remained stored even when power was removed. This was a critical advantage over previous volatile technologies. It was also more robust than tube memory.
• Development: Developed by Frederick Viehe and An Wang, and later refined by Jay Forrester and Jan Rajchman. First commercialized in the Whirlwind I computer (1953).

The Dawn of Semiconductor Memory (1960s)

The invention and development of transistors and integrated circuits (ICs) eventually led to the creation of memory chips made from semiconductor materials.

• Early Semiconductor Memory (Early 1960s): Initially implemented using discrete bipolar transistors as flip-flop circuits. These were faster than core memory but expensive and not very dense.
• First Semiconductor Memory IC: IBM introduced the SP95 chip in 1965, one of the first bipolar semiconductor memory ICs.

The MOS Memory Revolution (Late 1960s - 1970s)

The invention of the MOSFET (Metal-Oxide-Semiconductor Field-Effect Transistor) proved to be the key technology for creating dense, low-power, and cost-effective semiconductor memory.

MOSFET (Metal-Oxide-Semiconductor Field-Effect Transistor)

A fundamental building block of modern digital electronics. In memory, MOS transistors act as switches to control access to storage elements (like capacitors or flip-flops).

• Development: MOS memory was developed by John Schmidt at Fairchild Semiconductor in 1964. Improvements in MOS IC technology, like the silicon-gate process developed by Federico Faggin in 1968, enabled mass production.
• Impact: MOS memory offered higher performance, lower power consumption, and lower cost per bit compared to magnetic core memory. It quickly overtook core memory as the dominant technology in the early 1970s.
• Birth of SRAM and DRAM: This era saw the development of the two main types of volatile RAM:
  • SRAM (Static RAM): Robert Norman at Fairchild (1963). Later MOS SRAM by John Schmidt (1964). Relies on flip-flop circuits (typically 6 transistors per bit).
  • DRAM (Dynamic RAM): Robert H. Dennard at IBM (1966-1967) invented the single-transistor DRAM cell based on MOS technology (1 transistor + 1 capacitor per bit). This paved the way for much higher densities. The first commercial DRAM chip was the Intel 1103 (1970).
• Non-Volatile Semiconductor Memory: Development also proceeded for non-volatile semiconductor memory, leading to:
  • PROM (Programmable ROM): Wen Tsing Chow (1956). One-time programmable.
  • EPROM (Erasable PROM): Dov Frohman at Intel (1971), based on the floating gate concept by Kahng and Sze (1967). Erasable using UV light.
  • EEPROM (Electrically Erasable PROM): Yasuo Tarui et al. (1972). Electrically erasable, byte by byte.
  • Flash Memory: Fujio Masuoka at Toshiba (early 1980s). Electrically erasable in blocks, leading to higher density and lower cost than EEPROM.

This historical journey highlights the key properties that designers optimized for: speed, density, cost, volatility, and random access capability. Modern computers are built upon the foundation laid by the MOS semiconductor revolution.

Modern Semiconductor Memory Types

Modern computer memory primarily falls into two categories based on whether they retain data when power is removed: volatile and non-volatile.

Volatile Memory

Volatile Memory

Computer memory that requires power to maintain the stored information. When power is removed, the data is lost.

Volatile memory is typically used for primary storage (main memory) and CPU caches because it offers very high speed, which is crucial for the processor's operation.

• Static Random-Access Memory (SRAM):

  • Structure: Typically uses a circuit with 6 transistors per memory cell, configured as a flip-flop. This circuit holds its state (0 or 1) as long as power is supplied, without needing constant refreshing.
  • Characteristics: Very fast access times. Relatively low density compared to DRAM. More expensive per bit. Consumes more power when idle compared to DRAM (though less when actively writing).
  • Use Cases: Primarily used for CPU cache memory (L1, L2, L3 cache) due to its speed. Also found in smaller quantities in specialized applications or embedded systems where speed is paramount and capacity is small.

• Dynamic Random-Access Memory (DRAM):

  • Structure: Uses a single transistor and a capacitor per memory cell. The capacitor stores the charge representing a bit (charged = 1, discharged = 0).
  • Characteristics: Denser than SRAM (fits more bits in the same area), making it cheaper per bit. However, the charge on the capacitor leaks over time, so the memory needs to be periodically "refreshed" (read and rewritten) to prevent data loss. This refresh process makes it slower than SRAM. Requires more complex control circuitry.
  • Use Cases: The dominant technology for main system memory (RAM) in personal computers, servers, and mobile devices due to its high density and lower cost per bit.
  • Evolution: Modern main memory is typically Synchronous DRAM (SDRAM), which synchronizes its operation with the CPU clock for improved performance. DDR (Double Data Rate) SDRAM is a common type that transfers data on both the rising and falling edges of the clock signal.

Non-Volatile Memory

Non-Volatile Memory

Computer memory that can retain stored information even when power is removed.

Non-volatile memory is used for long-term storage (mass storage) and for storing data or programs that must persist even when the computer is off.

• Read-Only Memory (ROM):

  • Original Concept: Traditionally, ROM was programmed during manufacturing and could not be altered. Used for firmware or bootstrap code needed to start the computer.
  • Variations (Historical/Transitional):
    • PROM (Programmable ROM): Could be programmed once after manufacturing using a special device (a "PROM burner").
    • EPROM (Erasable PROM): Could be erased by exposure to strong ultraviolet light through a quartz window on the chip, and then reprogrammed.
    • EEPROM (Electrically Erasable PROM): Can be erased and reprogrammed electrically, byte by byte. Much more convenient than EPROM but often slower and more complex.

• Flash Memory:

  • Structure: A type of EEPROM that allows entire blocks of data to be erased and rewritten at once, rather than byte by byte. Uses floating-gate transistors.
  • Characteristics: High density, relatively low cost, reasonable speed (though slower than DRAM, faster than traditional magnetic storage). Has a limited number of erase/write cycles before wear-out occurs.
  • Types: NOR flash (faster read access, byte addressable, used for firmware/code storage) and NAND flash (higher density, faster write/erase, block addressable, used for mass storage). Flash memory can also use Multi-Level Cells (MLC) to store more than one bit per cell, increasing density but potentially reducing speed and endurance.
  • Use Cases: The backbone of modern solid-state drives (SSDs), USB flash drives, memory cards (SD cards), and storing firmware (like the BIOS/UEFI on a motherboard).

• Other Non-Volatile Technologies: While the above are common, historically, other non-volatile methods included magnetic tape, punched cards, floppy disks, hard disk drives (HDDs), optical discs (CDs, DVDs), and magnetic drums. These are generally considered mass storage rather than main memory, but they are non-volatile.

Semi-Volatile Memory

Semi-Volatile Memory

Memory that exhibits characteristics of both volatile and non-volatile memory, typically retaining data for a limited time after power loss or having a volatile mode used to extend the lifespan of the non-volatile elements.

This is a less common category but relevant in advanced memory system design. It describes technologies or configurations that aim to balance the speed and endurance of volatile memory with some degree of non-volatility.

• Concept: Often involves using volatile cells for high-speed writes, periodically copying data to a non-volatile section before power loss or to manage wear. Or, the non-volatile retention time might be limited, requiring specific power-down procedures.
• Examples: Some advanced memory technologies under development (like STT-MRAM) can operate in a semi-volatile mode (small, fast cells with limited retention) vs. a truly non-volatile mode (larger, slower cells with infinite retention). Another form is nvSRAM (non-volatile SRAM), which pairs SRAM with a small non-volatile memory on the same chip. Data is copied to the non-volatile part either on command or automatically on power loss (if power fails slowly enough). Battery-backed RAM is another example, where a battery provides power to volatile RAM during main power loss, making it semi-volatile based on battery life.

Memory Organization and Access

Regardless of the specific technology, memory is fundamentally organized to be accessed efficiently by the CPU.

• Addressability: Memory is addressable, meaning each location (word or byte) can be uniquely identified by an address.
• Address Bus: The CPU uses an address bus to send the address of the location it wants to access. The width of the address bus determines the maximum amount of memory the CPU can directly address (e.g., a 32-bit address bus can address 2^32 locations).
• Data Bus: The CPU uses a data bus to send data to memory (write operation) or receive data from memory (read operation). The width of the data bus is often related to the word size of the computer.
• Control Signals: Additional signals (like Read Enable, Write Enable, Chip Select) control the operation of the memory device, telling it whether to read or write and when to perform the action.

When building a system from scratch, understanding how to interface the CPU's address, data, and control buses to the memory chips is crucial. This involves memory decoding (using logic gates to select the correct memory chip based on the address range) and timing considerations to ensure data is stable when read or written.

Memory Management: The Role of the Operating System

In modern computer systems, memory is a shared resource used by the operating system and multiple running programs. Managing this resource effectively is critical for stability, performance, and security. This is primarily the responsibility of the operating system, often with hardware assistance from the CPU's Memory Management Unit (MMU).

Virtual Memory

Virtual Memory

A memory management technique implemented by the operating system, often with hardware support (MMU), that allows programs to use a larger address space than physically available in RAM. It uses secondary storage (like a hard drive) as an extension of RAM, swapping data between RAM and storage as needed.

• Concept: Programs operate using virtual addresses within their own isolated virtual address space. The MMU translates these virtual addresses into physical addresses in RAM. If the required data is not in RAM, the OS intercepts the request (a "page fault"), finds the data in a swap file on storage, loads it into RAM (potentially moving other data from RAM to storage to make space), updates the MMU's translation tables, and then allows the program to access the data.
• Benefits:
  • Allows programs to be larger than physical RAM.
  • Simplifies programming, as developers don't need to manage physical memory locations.
  • Enables multi-tasking, as the OS can manage memory for multiple programs simultaneously.
• Performance Impact: While beneficial, virtual memory involves transferring data between slow storage and fast RAM. If the system is constantly swapping data because there isn't enough RAM for all active processes, it leads to a condition called "thrashing," where the system spends more time swapping than doing useful work, severely degrading performance.

Protected Memory

Protected Memory

A system implemented by the operating system, again often with MMU hardware, that assigns each running program its own isolated region of memory and prevents it from accessing memory belonging to other programs or the operating system itself.

• Concept: The MMU enforces memory boundaries for each process. If a program attempts to read from or write to a memory address outside its assigned region, the MMU triggers an exception (like a "segmentation fault"), and the operating system typically terminates the offending program.
• Benefits:
  • Stability: A bug in one program is less likely to crash the entire system by corrupting the memory of other programs or the OS kernel. Only the faulty program crashes.
  • Security: Prevents malicious programs or code from accessing or altering sensitive data or code belonging to other processes or the operating system, helping to isolate security breaches.
• Contrast with Unprotected Memory: In systems without protected memory (like early operating systems or simple embedded systems), a single programming error (e.g., writing past the end of an array) could overwrite critical data belonging to another program or the OS, leading to unpredictable behavior, crashes, or security vulnerabilities that are easy to exploit (e.g., buffer overflows).

Memory Bugs

Improper handling of memory, particularly in languages like C or C++ where memory is managed manually, is a significant source of bugs and security vulnerabilities.

• Memory Leak: Occurs when a program allocates memory from the system but fails to free or deallocate it when it's no longer needed. Over time, the program's memory usage grows, potentially consuming all available memory and causing the program or even the entire system to run out of resources and crash.
• Segmentation Fault (Segfault): Happens when a program attempts to access a memory location that it does not have permission to access (often outside its allocated segment or trying to write to a read-only area). The operating system detects this violation via the MMU and terminates the program.
• Buffer Overflow: Occurs when a program writes data beyond the allocated boundary of a buffer (a block of memory). This overwrites adjacent memory locations, which might contain other data, program code, or control information. This can lead to erratic program behavior, crashes, or be maliciously exploited to execute arbitrary code.

While memory management concepts like virtual memory and protected memory are OS-level, understanding them is crucial for system builders. It highlights the architectural features needed (like an MMU) and the complexities that abstract away the raw hardware interaction for application programmers.

Beyond Main Memory

While this resource focuses on memory as primary storage, related concepts are vital for a complete picture:

• Memory Hierarchy: Modern systems use a hierarchy of memory types based on speed and cost. Faster, more expensive memory (like CPU cache - SRAM) is closer to the CPU, while slower, cheaper memory (like main RAM - DRAM, and storage - Flash/HDD) is further away but provides higher capacity. Data is moved between levels as needed.
• Processor Registers: Small, very fast storage locations directly within the CPU. They hold data and instructions currently being processed. While they store data, they are usually not considered "memory" in the same sense as RAM because they lack an addressing mechanism in the way main memory does and typically only hold a few "words" of data.

Understanding computer memory, from its historical roots in vacuum tubes and magnetic cores to the intricate semiconductor chips of today and the complex management techniques employed by operating systems, is essential for anyone seeking to truly grasp how computers work at a fundamental level or attempting to build or restore vintage systems. It reveals the clever engineering solutions developed to overcome limitations in speed, capacity, and cost, and the ongoing challenge of managing this vital resource.