IEEE Solid-State Circuits Magazine - Spring 2016 - 20

Figure 7: Magnetic core memory. (Photo
courtesy of the Charles Babbage Institute,
University of Minnesota.)

Advanced Study (IAS), RCA began
development of the Selectron tube for
the IAS computer. Originally targeted
to deliver 4,000 b of memory, it never
achieved this density; instead, it was

between 1947 and 1952 (Figure 7).
Implementation honors go to Jay Forrester of MIT, who had grown disillusioned with the use of unreliable
MIT dual-gun electron tube memory
for the MIT Whirlwind computer. The
initial core memory subsystems had a
density of 2,000 B and were installed
in the Whirlwind in 1953. The principal advantages of core memory were
the improved reliability over the electron tube memories and the random
access performance of 5 ns versus
the tube access time of 25 ns. Like
several of the preceding memories,
the read of magnetic core memory is
destructive, requiring re-writing of
the data after a read. The initial cost
of magnetic core memory was about
US$1.00/b and eventually cost about

If we exclude papyrus, paper, scribes, monks,
Gutenberg, and their cellulose-centered media
from consideration, modern information storage
may be traced to the invention by Danish
engineer Valdemar Poulsen of a magnetic wire
recording system in 1899.
finally commercialized at 256 b. The
advantage of the Selectron tube over
the Williams tube was that the read
was nondestructive and had a longer
persistence, which reduced much of
the need for refresh. Physically, the
Selectron tube moved the read plate
into the tube where alignment could
be ensured, thus resulting in higher
reliability than the Williams tube.
The complexities of the Selectron
tube design resulted in additional
expense, with production eventually
yielding parts at prices of US$500 for
256 b. Having missed its opportunity
to be used in Princeton's IAS machine,
the Selectron was finally used in the
RAND Corporation's JOHNNIAC.

Core Matters
Several inventors have their own
claims to magnetic core memory, which
was developed and commercialized

S P R I N G 2 0 16

US$0.01/b [5]. The reliability and
low cost of magnetic core memory
led to its near universal adoption as
computer main memory by the early
1960s, until semiconductor memory
emerged in the 1970s. In a nod to new
applications, magnetic core memory
was also adopted for use in noncomputing products, including the Seeburg V-200 jukebox. Now a collector's
item [6], these jukeboxes command a
nice price-if you can find one!
If magnetic core memory was
cementing its position in the core of
the computer (somewhat temporarily), other advances in memory were
occurring outside the CPU. For the first
time, distinctions were being made in
the memories attached to computers. Tape drives and hard drives were
solidifying their positions as "storage" devices as opposed to "memory"
devices. Hard-drive technology came

IEEE SOLID-STATE CIRCUITS MAGAZINE

to the market in 1956, with the IBM
350 RAMAC, a 50-disk, 1-ton behemoth with a capacity of 3.75 MB. It had
an average access time of about 600 ns,
an areal density of about 2,000 b/in2,
and a cost of about US$9,000 per MB [7].

DRAM Dynasty
In 1966, Dr. Robert Dennard at the
IBM Thomas J. Watson Research Center invented the "field effect transistor memory," now known as DRAM.
With one transistor and one capacitor per cell, and built on a semiconductor process, it had the low-cost
and high-performance characteristics necessary to find its way into
computing systems. Like capacitive
drum and Williams tube memories
before it, DRAM needs periodic
refresh and is volatile on power loss.
DRAM was not an immediate success,
however. It was not until 1970, when
Intel introduced the revolutionary
1103 DRAM (a three-transistor-percell design), that the industry began
to take DRAM seriously. Large numbers of the 1103 didn't become available until 1971. In 1973, the Mostek
Corporation introduced the MK4096,
a 4,000-b design. Mostek followed
this in 1976 with a design using the
1967 IBM patent to build a 16,000-b,
1-T cell MK4116. The race for DRAM
density was on.
Early DRAM devices used planar
capacitors for charge storage. As
the quest for density began to push
cells into tighter X-Y dimensions,
the Z-axis became a differentiating
path for DRAM players. Some companies chose to etch capacitors deep
into the silicon-the so-called deep
trench capacitors. Other companies
chose to build capacitors on top of
the silicon-the stacked capacitors.
Today, all high-volume commercial
DRAM devices are built with the
stacked-capacitor technology, while
trench capacitors remain in use for
certain embedded DRAM processes.
The advent of a reliable, fast, and
inexpensive main memory technology truly enabled the personal
computer revolution. As processor
speeds advanced with each generation,

Table of Contents for the Digital Edition of IEEE Solid-State Circuits Magazine - Spring 2016